1.1 Motivation

Vehicle Ad Hoc Networks (VANETs) are becoming more prevalent as a fundamental component of Intelligent Transportation Systems (ITS) [1,24]. When vehicles share information in real-time, it is essential to make sure it is trusted in order to prevent malicious behavior, misinformation, and cyberattacks that could compromise road safety. VANETs are highly dynamic and decentralized, so traditional centralized trust management is insufficient. As a result, trust relationships are volatile, and single points of failure are increased [17,25].

Our design is motivated by the need for decentralized, tamper-resistant, and auditable trust mechanisms in dynamic vehicular environments, requirements that traditional reputation systems often fail to meet due to their reliance on fixed infrastructure or centralized trust anchors [26,27]. Decentralized and immutable blockchain technology offers a promising solution to these problems [22,28,29]. In addition, the use of traditional consensus mechanisms such as Proof of Work (PoW), increases network overhead, computational burden, and latency. In order to leverage blockchain strengths while mitigating its weaknesses, it is imperative to explore more efficient trust management frameworks.

To address the critical issues of decreasing network overhead, reducing computational and energy costs, and reducing latency while maintaining VANET communications security and reliability, an optimized blockchain-based trust management model was developed in this study [4,10,30]. This research aims to contribute to the broader effort of enhancing blockchain technology viability in VANETs by integrating lightweight consensus mechanisms with efficient data management strategies.

1.2 Contributions

The primary contributions of this paper are outlined as follows:

• Optimized Blockchain-Based Trust Management Framework: To manage trust specifically in Vehicular Ad Hoc Networks (VANETs), the proposed framework utilizes blockchain technology. To improve network performance in high-mobility vehicular environments, the framework integrates lightweight consensus mechanisms such as Practical Byzantine Fault Tolerance (PBFT) and Proof of Stake (PoS).
• Decentralized Trust Evaluation with Reduced Network Overhead: A decentralized trust evaluation model is proposed, effectively distributing trust management processes across multiple nodes. Decentralization reduces the risk of single points of failure by eliminating dependency on a central authority. Furthermore, the model introduces a mechanism for selectively propagating trust-related data by only distributing it to nodes directly involved in the transaction or interaction.
• Energy-Efficient Trust Management: The proposed framework incorporates an energy-aware trust management protocol to address energy consumption challenges in resource-constrained vehicular environments. Using this protocol, participants in the blockchain network are dynamically adjusted according to their energy resources, thus extending the network’s lifetime while maintaining high trust accuracy.
• Integration of Smart Contracts for Automated Trust Management: Blockchain-based trust management uses smart contracts for automating key trust management processes, such as computing, validating, and disseminating trust scores. Through smart contracts, trust management can be enhanced and made more reliable, while also reducing the number of communications between nodes, thereby reducing network overhead.

1.3 Organization of the paper

The remainder of this paper is organized as follows. Sect 2 presents a comprehensive review of the state-of-the-art in trust management for VANETs, with a particular emphasis on blockchain-integrated approaches. This section identifies key research challenges and motivates the need for decentralized, tamper-proof trust frameworks. Sect 2 introduces the proposed blockchain-based decentralized trust management architecture. It outlines the system model, including the roles of On-Board Units (OBUs), Road Side Units (RSUs), and the Trust Authority (TA), and describes the mechanisms for vehicle authentication, trust evaluation, and selective message propagation.

Sect 3 details the technical foundation of the proposed framework, including the vehicle-to-vehicle authentication algorithm and the hybrid Proof-of-Stake (PoS) and Practical Byzantine Fault Tolerance (PBFT) consensus mechanism. The section formalizes trust computation strategies and block validation protocols under vehicular mobility constraints.

Sect 5 provides a thorough performance evaluation of the framework using simulations based on real-world VANET datasets. Key performance indicators such as trust accuracy, propagation latency, consensus overhead, scalability, and resilience against malicious behaviors are systematically analyzed.

Finally, Sect 7 concludes the paper by summarizing the contributions, emphasizing the significance of the proposed solution in addressing trust challenges in VANETs, and outlining potential directions for future work in the domain of secure vehicular communication systems.

2.1 Blockchain-based trust management models

Blockchain technology has recently gained traction as a potential solution for overcoming the limitations of both centralized and distributed trust management models in VANETs [18]. The decentralized, transparent, and immutable characteristics of blockchain make it well-suited for securing communication in highly dynamic and distributed environments like VANETs. In blockchain-based trust management models, trust scores and related information are recorded on a distributed ledger, ensuring that data is tamper-proof and verifiable by all participating nodes [11]. Blockchain technology has emerged as a promising solution to the limitations of both centralized and distributed trust management models. Each vehicle or roadside unit in the network can participate in the consensus process to validate trust data, ensuring that the trust scores are accurate and have not been manipulated. The decentralized nature of blockchain also removes the reliance on a central trust manager, eliminating the risk of single points of failure and improving the overall security of the trust management system [43].

Several scholars have proposed blockchain-based trust management models specifically tailored for VANETs. In [44], a hybrid trust management model based on blockchain was proposed to combine vehicle-based trust calculations with blockchain consensus mechanisms. This approach ensures that trust scores are locally calculated and globally verified, offering a robust solution for secure communication in VANET environments. In [45], the authors propose a blockchain-based decentralized trust management system for VANETs that leverages a hybrid consensus mechanism combining Proof of Stake (PoS) and Practical Byzantine Fault Tolerance (PBFT). The system achieves high security and reduces computational overhead by utilizing PBFT for fast consensus and PoS to maintain decentralization. However, the hybrid model introduces challenges in terms of scalability, as the PoS mechanism requires nodes to validate transactions, which can become resource-intensive in large networks. [46] presents a blockchain-based trust management framework that focuses on privacy-preserving communication in VANETs. The model uses zero-knowledge proofs to verify vehicle identities without revealing personal information to the network. While this approach improves privacy and security, it introduces significant computational overhead during the verification process. The time delay caused by these cryptographic techniques can hinder real-time trust management in fast-moving vehicular networks. In [47] proposes a lightweight blockchain-based trust management system designed to reduce the computational burden typically associated with blockchain technology. The model integrates a lightweight consensus algorithm that minimizes the energy consumption and latency of blockchain operations. Although the model significantly improves computational efficiency, it does so at the cost of reduced security guarantees compared to more robust consensus mechanisms like Proof of Work (PoW) or PoS. The lightweight consensus may not be resilient against sophisticated attacks such as double-spending or Sybil attacks, particularly in highly dynamic environments like VANETs [27].

In [48], the authors implement a blockchain-based trust management model that combines vehicle-to-vehicle (V2V) communication with blockchain-based trust validation. Each vehicle calculates a local trust score and periodically submits the score to the blockchain, where it is validated through a consensus process. While this model enhances the scalability of the trust management process by distributing the computational load across multiple vehicles, it introduces communication overhead due to frequent interactions between vehicles and the blockchain network. The increased communication can lead to network congestion, particularly in dense traffic scenarios, thus reducing overall system performance.

3.1 System components

The proposed system architecture consists of three main components: On-Board Units (OBUs) installed in vehicles, Roadside Units (RSUs) deployed along transportation infrastructure, and a Trust Authority (TA) that oversees network initialization and credential issuance. The TA is responsible for registering entities, issuing cryptographic credentials, and setting global system parameters to ensure secure and verifiable identities prior to vehicles joining the network.

Importantly, the blockchain network, not the Trust Authority (TA), is responsible for decentralized trust evaluation, consensus, and ledger maintenance. The TA does not participate in operational trust computation or block validation; these functions are fully carried out by distributed validator nodes using the hybrid PoS PBFT mechanism. This separation of responsibilities ensures that while the TA provides regulatory oversight and initialization, the trust management process itself remains fully decentralized and tamper resistant.

Vehicles communicate through Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) links to exchange information and generate trust scores based on direct and indirect interactions. All validated data are recorded on the blockchain ledger, enabling transparent, auditable, and secure trust management without centralized control.

3.2 Trust Authority (TA)

The Trust Authority (TA) plays a critical supervisory and bootstrapping role in our decentralized trust management system. While the TA is assumed to be secure and reliable, as in similar works [33], it does not act as a centralized controller. Instead, it facilitates decentralized trust operations in VANETs by overseeing credential issuance, validator selection, and trust bootstrap processes. In our architecture, the TA assists in initializing the network by securely registering OBUs and RSUs, issuing cryptographic credentials, and managing the setup of validator nodes. Once initialization is complete, trust evaluations, block generation, and consensus operations are entirely decentralized and carried out through validator nodes using the hybrid PoS–PBFT mechanism.

Furthermore, the TA maintains visibility over vehicle data and supports real-time identification of potentially malicious entities [49]. It collects the trust scores, , for each vehicle over time t, where the scores are derived from both direct and indirect interactions as follows:

(1)

where represents the direct trust score based on prior behavior, and denotes the recommendation of node regarding . Each recommendation is weighted by w_j, which reflects the recommender’s trust level and is computed as a normalized trust value over recent interactions:

(2)

The parameters α and β are weighting coefficients with , and n is the number of neighboring nodes. For newly joined vehicles, the TA provides an initial trust value to establish a starting reputation profile. This initialization ensures that new participants can engage in the network immediately while their trust scores are continuously updated through decentralized evaluation and consensus as they interact with other nodes.

As vehicular dynamics evolve, trust updates remain responsive yet stable through exponentially weighted moving averages (EWMA) in both and . Importantly, these initialization and monitoring processes do not interfere with or influence consensus decisions. All trust updates are recorded on-chain and validated through the distributed validator network.

The rationale for incorporating a TA is grounded in practical deployment scenarios, where regulatory or transportation authorities require auditability and accountability in safety-critical networks like VANETs. The TA may issue revocations and compliance notices; however, it possesses no decision-making authority over consensus or trust score updates. Even if the TA were compromised, it could not retroactively alter trust records or control consensus, since these functions are governed by the blockchain network.

3.2.1 Convergence analysis and TRF threshold selection.

This subsection provides (i) a brief convergence analysis for the EWMA updates used in direct and recommendation-based trust components, and (ii) a principled procedure for selecting the Trust Relevance Factor (TRF) threshold used in selective propagation.

Let the direct-trust update follow the EWMA recursion

(3)

where is the smoothing parameter and I_t is the interaction outcome at time t. Assume {I_t} is a stationary sequence with mean and variance . Taking expectations yields

(4)

so that,

For the variance, assuming independence between innovation terms for simplicity, the steady-state variance of D_t is

(5)

Thus the EWMA estimator is unbiased in the limit and its variance is bounded and controlled by λ. Smaller λ yields smoother but slower adaptation; larger λ increases responsiveness at the cost of higher steady-state variance.

3.2.1.1 Boundedness and convergence of the composite trust score.

The composite trust score is given by

(6)

with , , and normalized weights satisfying . If each component D_t and R_j,t is bounded in [0,1], then for all t. Under mild ergodicity conditions on the underlying interaction/recommendation processes, each component converges in expectation to its long-run mean and therefore T_t converges in expectation to a convex combination of these means. Because the weights w_j,t are normalized and derived from past trust values, they remain bounded and do not induce divergence.

3.2.1.2 Definition and normalization of TRF.

Recall TRF between nodes and is defined as

(7)

where F_ij is the interaction frequency (normalized to [0,1] over a sliding window) and I_ij is the importance metric (also normalized to [0,1]). For clarity and stability we recommend imposing and ; normalization of F_ij and I_ij should be performed by min–max or by dividing by the maximum observed value over the calibration window.

3.2.1.3 Principled TRF threshold selection.

We recommend a hybrid, reproducible approach comprising three options depending on the availability of labeled calibration data and deployment constraints.

Option A (ROC-based calibration, preferred when labeled calibration data are available). Let τ denote the TRF threshold and define the detection rule “propagate if ”. Compute true positive rate TPR and false positive rate FPR over a labeled calibration set. Choose

(8)

i.e., maximize Youden’s J statistic to obtain the operating point with the best trade-off between sensitivity and specificity. Alternatively, select τ to satisfy a deployment-specific constraint on FPR.

Option B (statistical adaptive rule, for online operation). Maintain a rolling window of the most recent W TRF samples among relevant neighbors, compute the rolling mean and standard deviation , and set

(9)

where k is chosen to control the Type I error (e.g., k = 1.96 for a one-sided 97.5% threshold under approximate normality). This rule adapts to local traffic density and behavior.

Option C (percentile-based fallback, for deployments without labeled data). Compute the empirical p-th percentile Q_p of TRF over a calibration period and set with p in [0.70,0.90] depending on desired sensitivity.

For typical vehicular network deployments, we recommend the following initial parameter settings: an EWMA smoothing factor , where smaller values yield smoother but slower trust updates; a TRF importance weight , with used in our baseline experiments to emphasize message criticality; and a calibration window W chosen based on expected neighbor churn. These parameters should subsequently be validated and fine-tuned using real deployment traces or the VeReMi dataset through the ROC-based calibration procedure described above.

3.3 Blockchain-enabled trust management and selective propagation

The proposed VANET trust management framework leverages blockchain to ensure a decentralized, secure, and tamper-resistant infrastructure for trust evaluation and enforcement. At the core of this architecture, each vehicle is equipped with an On-Board Unit (OBU), enabling real-time communication with other vehicles (V2V) and Roadside Units (RSUs) through Vehicle-to-Infrastructure (V2I) communication. RSUs act as relay nodes that assist in transmitting trust-related data to the blockchain network, forming the backbone of secure vehicular communication. Fig 2 illustrates the overall architecture, showcasing the interplay between OBUs, RSUs, the TA, and the blockchain infrastructure. Vehicles exchange trust-relevant information during mobility, which is then filtered, validated, and securely stored in the blockchain. Smart contracts enforce automatic trust updates and policies, ensuring system autonomy and eliminating the need for centralized control. The proposed system integrates vehicular communication, trust evaluation, consensus validation, blockchain-based recordkeeping, and selective data dissemination to provide a robust, decentralized trust management framework tailored for VANET environments.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Proposed blockchain-based VANET system model.

The architecture consists of On-Board Units (OBUs), Roadside Units (RSUs), and a Trust Authority (TA) with a bootstrapping and supervisory role for credential issuance, identity registration, and validator setup. The TA does not participate in trust computation, consensus, or ledger maintenance. Vehicles communicate via V2V and V2I links, while trust scores are derived from direct interactions and peer recommendations. All validated data are stored on a tamper-proof blockchain maintained by distributed validators, ensuring decentralized, auditable, and secure trust management with regulatory oversight.

https://doi.org/10.1371/journal.pone.0338618.g002

Trust computation is coordinated by a semi-centralized Trust Authority (TA), which plays a supervisory but non-authoritative role. It aggregates direct interaction outcomes and peer feedback to generate a dynamic trust score for each vehicle, as formalized in Eq (1). The TA also assists in identifying misbehaving entities, ensuring that the reputation system remains adaptive and responsive. However, the trust evaluation outcomes are not finalized until they pass through a decentralized consensus process, ensuring resilience against central point failures. The trust scores and validated vehicular interactions are transmitted to a blockchain layer, where they are permanently recorded in a transparent, immutable ledger. This blockchain ledger serves as the system’s trust repository, storing all verified transactions in sequential blocks (Block 1, Block 2, etc.), ensuring auditability and historical traceability. These blocks form a cryptographically linked chain that supports tamper-proof recordkeeping of vehicular trust histories. Consensus on the validity of transactions is achieved through a hybrid Proof of Stake (PoS) and Practical Byzantine Fault Tolerance (PBFT) mechanism. Consensus validators are selected based on accumulated trust stakes, promoting fairness and energy efficiency through PoS, while PBFT ensures fault tolerance and rapid finality by requiring agreement among a majority of trustworthy nodes. This hybrid consensus framework maintains high security and performance, making it suitable for high-mobility vehicular environments.

To avoid overwhelming the network with redundant updates, a selective data propagation mechanism is incorporated. Rather than broadcasting all trust-related data globally, only contextually relevant updates are disseminated to nodes that have had recent or meaningful interactions with the evaluated vehicle [50]. This targeted approach significantly reduces communication overhead and enhances system scalability without compromising trust transparency.

3.4 Blockchain-based trust management scheme

This section presents our proposed trust management scheme in detail, which comprises four parts, namely, system initialization, vehicle credibility evaluation, event credibility evaluation, and storage of indirect trust through blockchain. The comprehensive explanation of our suggested scheme can be stated as follows.

3.5 System initialization

• TA defines an elliptic curve , where G is a finite cyclic group of order q on the elliptic curve, q is a large prime number, P is the generator of the group G, and , with . Then, TA exposes these parameters to all nodes.
• To join the network, vehicle forwards a registration request to the RSU, generating a random number S_ki as its private key and producing its public key . further generates its certificate utilizing the private key, which is then transmitted along with the public key to the RSU.
• Upon receiving the registration request, the RSU stores the vehicle’s certificate and the indirect trust IT_i = 0 on the blockchain. The RSU deploys smart contracts and utilizes the vehicle’s public key as an identifier to retrieve certificates and trust values, reducing the storage burden on the vehicle. It is assumed that the RSU entities are trustworthy. Additionally, Bayesian classifiers trained on the VeReMi dataset are used to enable vehicles to compute the direct trust of the target vehicle.

3.6 Evaluation of vehicle credibility

Evaluating the credibility of vehicles is a crucial part of the trust management process. This assessment involves calculating the vehicle trust value by combining direct and indirect trust. Direct trust refers to the credibility established through direct observations or interactions, while indirect trust is derived from evaluations made by other vehicles.

1) Direct trust calculation: Direct trust is calculated using a GaussianNB classifier, which is trained on the refined VeReMi dataset. For each vehicle, its state information S is input to the Bayesian classifier, which computes the direct trust DT_i. The specific process for computing the direct trust will be discussed in the “Trust Value Calculation” section.

2) Indirect trust calculation: The indirect trust IT_i of a newly registered vehicle is initialized to zero. Our method employs an active detection scheme where neighboring vehicles and the RSU work collaboratively to detect the trust level of the target vehicle, as shown in Eq 10. The neighbor vehicle sends a test message to vehicle , requesting it to forward the message to the RSU. The RSU compares the message received from and . If both are identical, is considered legitimate, and a trust score T_i = 1 is assigned. If not, is deemed malicious, and a score of T_i = 0 is sent to the TA. The indirect trust IT_i is computed as:

(10)

3) Updating Indirect Trust: For already registered vehicles, the indirect trust IT_i is updated using the formula:

(11)

Here in Eq 11, is a weight parameter that controls the rate of change in indirect trust. This allows for dynamic adjustment during the detection of malicious vehicles. The updated trust values are stored on the blockchain via the RSU using a consensus algorithm.

4) Trust Value Calculation: The overall trust value for a vehicle is calculated by combining direct trust DT_i and indirect trust IT_i is computed in Eq 12:

(12)

The trust value must exceed a threshold for the vehicle to be considered credible.

3.7 Evaluation of event credibility

In addition to vehicle trust, assessing the credibility of events is essential to prevent the dissemination of false information, which can lead to serious consequences in VANETs [51]. The event credibility evaluation process ensures that the events reported by vehicles are verified before being accepted and broadcasted to the network. The steps involved in this process are described below.

1) Event Reporting by Vehicle: When a vehicle detects an event E such as a traffic accident or hazard, it broadcasts the event information to surrounding vehicles and requests these vehicles to provide their judgment on the event.

2) Evaluation Accuracy Calculation: The accuracy of the information provided by the vehicle can vary depending on several factors, including the trust level of the vehicle and its distance from the event location. To account for these factors, each surrounding vehicle computes its evaluation accuracy C_j using the following formula:

(13)

Here in Eq 13, is the trust value of vehicle , d_j is the distance between vehicle and the location of the event E, e is Euler constant, and is a weight parameter that adjusts the impact of the distance on the evaluation accuracy. By adjusting , the system can modify the degree to which the distance affects the credibility of the event.

Fig 3 presents a blockchain-based model for event credibility evaluation in a VANET system. When an event occurs, such as an accident or hazard, a vehicle initiates a message broadcast with relevant event data, including context C_i and transmission rate . Nearby vehicles, represented as the set , receive the message and send verification requests to the Road Side Unit (RSU). The RSU aggregates verification requests and calculates credibility scores based on factors such as vehicle context and message integrity, formalized as . The RSU then initiates a consensus process, denoted as , where multiple RSUs collaborate to confirm the event’s credibility. The consensus results are recorded in a new block B_t that is subsequently stored in the blockchain ledger , ensuring that verified event data remains immutable and tamper-proof. This model enhances data integrity and trust in VANET communications by securely evaluating event credibility.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Blockchain-based model for event credibility evaluation in VANETs, utilizing RSU consensus to secure and verify event data on the blockchain.

https://doi.org/10.1371/journal.pone.0338618.g003

3) Judgment of Event: After calculating its evaluation accuracy, each vehicle provides its judgment of the event, denoted as for a positive judgment (indicating that the event is valid) or for a negative judgment (indicating that the event is invalid). These judgments are then sent back to the vehicle that initially reported the event.

4) Calculation of Event Probability by RSU: Vehicle aggregates the judgments from surrounding vehicles and forwards the data to the nearest RSU. The RSU calculates the probability of event E being true using the following equation:

(14)

In Eq 14, n represents the number of positive judgments received for the event, and m represents the number of negative judgments. The probability P(E) determines the likelihood of the event being credible based on the aggregated evaluations.

5) Broadcasting the Event: If the probability P(E) exceeds a predefined threshold, the event E is deemed credible and the RSU broadcasts the event to all vehicles in the network. The indirect trust of the participating vehicles is updated accordingly. Conversely, if P(E) falls below the threshold, the event is considered not credible, and the RSU refrains from broadcasting it. Vehicles that falsely reported the event may be penalized by reducing their trust value. The formulas for updating indirect trust are as follows:

(15)

Eqs 15 and 16, is a weight parameter that controls the rate of trust adjustment based on the vehicle judgment. For the vehicle that initially reported the event, the indirect trust is updated as:

(16)

These updates ensure that vehicles contributing accurate judgments maintain or improve their trust levels, while those providing inaccurate information experience a decrease in trust.

3.8 Decentralized trust evaluation

The proposed model evaluates trust in a decentralized manner, where each node independently calculates the trust score of its neighboring nodes based on both direct interactions and recommendations from other nodes. The total trust score of a node at time t is defined as a weighted combination of direct trust and reputation-based trust [52]:

(17)

where in Eq 17:

• is the direct trust score of node at time t, computed based on its interactions.
• represents the recommendation provided by node regarding at time t.
• n is the number of neighboring nodes that provide recommendations.
• w_j is the weight assigned to node ’s recommendation based on its historical reliability.
• α and β are weights that balance the contribution of direct trust and reputational trust, such that .

The direct trust is calculated based on node behavior during interactions, such as message authenticity, consistency, and adherence to protocol. It is updated over time based on exponential moving averages (EMA) of past interactions, providing a smoothed trust evaluation:

(18)

where in Eq 18:

• represents the interaction outcome at time t (i.e., a binary value indicating whether the interaction was trustworthy or not).
• λ is a smoothing factor that controls the influence of recent interactions over past ones.

In this decentralized model, each node autonomously updates its trust evaluations based on its observations and shared reputations, without relying on any centralized authority. This ensures scalability and resilience in dynamic network environments.

3.9 Selective data propagation

To address the communication overhead and scalability challenges inherent in VANETs, we introduce a selective data propagation mechanism driven by a Trust Relevance Factor (TRF). Unlike traditional gossip-based dissemination protocols which often suffer from excessive redundancy and delayed convergence in highly dynamic networks. Our approach employs a deterministic, relevance-aware strategy that selectively targets only those nodes with meaningful interaction histories.

The core idea hinges on the TRF metric, which quantifies the necessity of propagating trust updates based on both the frequency and criticality of interactions between nodes. Specifically, the TRF between nodes and is defined as follows:

(19)

Here, denotes the frequency of interactions within a sliding time window w, calculated as:

while captures the importance or criticality of exchanged information (e.g., safety or event-based messages). The weighting parameters γ and δ allow the system to tune the influence of interaction frequency and importance, respectively. Through empirical tuning, we set to balance detection sensitivity and false positives under mobility-induced uncertainty.

When the TRF exceeds a predefined threshold τ, a trust update is disseminated as:

(20)

To identify relevant peers for propagation, each vehicle maintains a dynamically updated neighbor table constructed from periodic beaconing specifically Basic Safety Messages (BSMs), standard in DSRC and C-V2X protocols. This passive discovery mechanism enables vehicles to track neighboring nodes in real-time and determine which among them require trust updates based on the TRF computation. The process is inherently localized, eliminating the need for network-wide broadcasts or global state awareness. By disseminating trust information solely to impacted neighbors, the proposed scheme significantly reduces redundant communication while ensuring that vital trust updates reach the appropriate nodes. This localized propagation approach ensures bandwidth efficiency and preserves system responsiveness even in densely populated or high-mobility vehicular environments. Moreover, in contrast to probabilistic gossip-based protocols, our TRF-guided approach is deterministic, yielding better convergence and lower communication overhead. As demonstrated in Sect 5, our mechanism achieves over 95% trust accuracy while reducing communication overhead by approximately 35%, outperforming conventional epidemic dissemination models.

Lastly, the proposed scheme operates within a decentralized trust management framework, validated using a hybrid consensus mechanism that combines Proof of Stake (PoS) and Practical Byzantine Fault Tolerance (PBFT). This ensures both the integrity and consistency of trust scores across the network, supporting secure and scalable trust validation without reliance on centralized infrastructure.

3.10 Enhanced hybrid consensus mechanism for VANETs: Mobility-aware PoS with adaptive PBFT

In dynamic vehicular environments, directly applying conventional consensus protocols such as PoS or PBFT faces fundamental challenges related to validator volatility, stake stability, and consensus liveness. To address these issues, we design a hybrid consensus mechanism that combines mobility-aware Proof of Stake (PoS) with adaptive Practical Byzantine Fault Tolerance (PBFT). Unlike previous approaches that simply merge these protocols, our design introduces validator selection and consensus adaptations tailored specifically for VANET mobility patterns.

3.10.1 Mobility-aware validator selection via PoS.

Traditional PoS mechanisms rely on static, coin-based stake accumulation, which is incompatible with the highly mobile and transient nature of vehicular networks. To overcome this, we define a dynamic trust stake that reflects the recent behavior and trustworthiness of each node:

(21)

where represents direct trust and denotes recommendations from neighboring nodes. Stake values are computed over a sliding time window, ensuring that only nodes with consistently trustworthy behavior are eligible for validator selection. Validators are chosen from a short-term validator pool composed of RSUs and low-mobility OBUs within a geofenced area. The validator selection probability is given by:

(22)

Re-selection occurs periodically (every 5–10 seconds), preventing transient nodes from accumulating stake over long periods and maintaining validator stability during high-speed mobility. This mobility-aware design effectively addresses the stake persistence issue in traditional PoS.

3.11 Storage of indirect trust through blockchain

The storage of vehicle certificates and indirect trust values on a blockchain is proposed to prevent malicious tampering of data and ensure secure storage.

Blockchain decentralized and immutable nature significantly enhances system security. However, the delay in updating trust values depends on the consensus algorithm utilized by the blockchain. The PBFT algorithm, though effective for networks with a smaller number of nodes, faces scalability issues as it requires pairwise communication among all nodes [38]. As the number of participating nodes increases, the communication overhead becomes prohibitive.

As shown in Fig 4, the hybrid consensus mechanism combines PoS for dynamic selection of validators and PBFT for achieving agreement on trust scores. The system recalculates PoS-based trust stakes periodically to account for the mobility of vehicles in VANETs, while PBFT rounds ensure low-latency, secure finalization of blocks. This approach balances scalability, security, and efficiency in managing indirect trust on the blockchain [38].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Flow of the proposed hybrid consensus mechanism integrating Proof of Stake (PoS) for dynamic validator selection and Practical Byzantine Fault Tolerance (PBFT) for trust score agreement.

The system adapts to VANET mobility by periodically recalculating PoS-based trust stakes and using low-latency PBFT rounds to finalize blocks securely and efficiently.

https://doi.org/10.1371/journal.pone.0338618.g004

Our system adopts the BFT-Delegated Proof of Stake (BFT-DPoS) consensus algorithm, which offers a more scalable alternative. In this approach, representative nodes are selected through a voting process to handle the validation and verification tasks. Upon receiving a notification of a new block, these representative nodes verify the integrity and legitimacy of the block’s transactions. Once a majority of the representative nodes confirm the block’s validity, it is appended to the blockchain.

The BFT-DPoS algorithm operates as follows: The system elects 21 nodes with the highest votes as super nodes. These super nodes then select a miner among themselves based on network resources. The miner broadcasts a signed request with a timestamp to all super nodes. Upon verifying the request, the super nodes audit the results and communicate their findings to other nodes. When the audit results from the super nodes are consistent, and the number of matching results exceeds 2f, where f represents the number of Byzantine nodes, a confirmation message is broadcast to all nodes. If the super node collects more than confirmations, the block is validated and added to the blockchain. If consensus is not reached, the miner initiates additional consensus rounds to resolve any discrepancies.

3.12 Network overhead minimization

The combination of selective data propagation and decentralized consensus significantly reduces network overhead. First, by limiting trust updates to only relevant nodes, the model minimizes unnecessary communication and bandwidth usage. Second, the PoS mechanism ensures that only a subset of nodes participate in the consensus, reducing the computational burden on the network. The total communication overhead O_comm can be modeled as:

(23)

where O_local represents the overhead for local trust updates, and O_propagate represents the overhead for propagating trust data to neighboring nodes when , as shown in Eq 23. By controlling O_propagate through selective propagation, the model ensures that communication costs remain low.

The proposed methodology provides a decentralized and efficient approach to trust management in VANETs. By using selective data propagation and decentralized consensus, the model achieves secure and scalable trust evaluations while minimizing network overhead. The mathematical formulation of trust evaluation, relevance factor, and consensus ensures that the system is both robust and computationally efficient.

3.13 Comparison with traditional PKI-based authentication

Traditional Public Key Infrastructure (PKI) mechanisms rely on certificate authorities (CAs) to validate the identity of nodes in vehicular networks. These systems offer low-latency cryptographic authentication and are effective in relatively stable environments. However, VANETs operate in dynamic, large-scale, and often adversarial conditions, where static certificate verification is insufficient to assess the ongoing behavior of participating vehicles.

In contrast, our proposed blockchain-based mechanism integrates identity verification with dynamic trust evaluation by invoking smart contracts. While this may introduce marginal latency due to blockchain queries, it ensures that both the identity and historical behavior (trust profile) of the sender vehicle are jointly evaluated. This dual-layer verification enhances security and context awareness, particularly in detecting on-off attackers and sybil nodes that may possess valid certificates but exhibit malicious behavior. Moreover, anchoring trust evaluations on-chain offers several advantages: (i) it enables tamper-resistant and auditable logging of trust decisions; (ii) it supports cross-domain trust interoperability without reliance on a centralized authority; and (iii) it facilitates decentralized revocation and trust updates in real-time. Lightweight query mechanisms and local caching at RSUs further minimize overhead and ensure scalability in practical deployments.

Therefore, while PKI mechanisms excel in static identity management, the proposed blockchain-enabled trust-aware authentication better aligns with the dynamic and adversarial nature of VANETs, offering a robust, decentralized alternative that goes beyond identity to include behavioral validation.

5.1 Simulation tools and environment

This section presents the experimental framework adopted to validate the proposed decentralized trust management mechanism for VANETs. The proposed trust management framework is designed primarily for small- to medium-scale VANET scenarios such as urban zones or localized vehicular regions. The simulation settings were chosen to allow controlled experimentation and repeatability while capturing the trust dynamics in decentralized environments. Large-scale deployment and real-world testing are identified as future work directions due to the infrastructural and cost constraints involved.

To evaluate the system under realistic traffic and communication dynamics, we employed an integrated co-simulation setup. Vehicular mobility was modeled using the Simulation of Urban MObility (SUMO), while message dissemination and protocol simulation were handled by OMNeT++ integrated through the Veins framework [53]. The trust evaluation logic, including the Gaussian Naive Bayes (GaussianNB) model and blockchain mechanisms, were implemented in Python 3.11. For distributed ledger validation, a lightweight prototype was deployed using Hyperledger Fabric v2.4. Scikit-learn was used for training and testing the machine learning classifiers.

5.2 Simulation scenario and parameters

The proposed approach was evaluated in a simulated VANET environment consisting of 100 vehicles dispersed across an area of 4 square kilometer. Vehicles moved with speeds ranging from 10 to 20 m/s and operated under a communication radius of 300 meters. Of these, 20 vehicles were configured as malicious actors, while 80 operated legitimately. Simulations captured trust value fluctuations for both legitimate and malicious vehicles. The effectiveness of the trust management framework was quantified using the detection rate of malicious nodes and compared against benchmark approaches.

For evaluating communication overhead reduction, the baseline model corresponds to a traditional broadcast-based trust dissemination approach, widely adopted in early VANET trust management schemes. In this baseline, each trust update is broadcast to all neighboring vehicles without selective filtering, hierarchical aggregation, or validator mediation. This results in a linear growth of message overhead with network size. The proposed model contrasts this by selectively propagating trust updates through validator nodes and zone-level filtering, thereby reducing redundant transmissions. This baseline model was explicitly implemented in the same simulation environment to ensure fair and controlled comparison.

Additionally, the performance of the V2V authentication mechanism was examined in terms of computational overhead, and the proposed hybrid consensus algorithm was evaluated relative to conventional alternatives. The simulation results provided a quantitative basis for asserting the superiority of the proposed trust mechanism in terms of accuracy, efficiency, and resistance to malicious manipulation [54,55].

5.3 Hardware and software configuration

The simulation platform was deployed on a high-performance workstation. Table 4 provides a concise overview of the hardware and software configuration used in the experiments.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Hardware and software configuration.

https://doi.org/10.1371/journal.pone.0338618.t004

To emulate a distributed and decentralized environment, multiple validator nodes were instantiated using Docker containers configured with realistic latency and packet loss constraints.

To support independent verification, the core Python simulation scripts and Solidity smart contracts developed for this study have been released as open-source in a public GitHub repository [56]. This repository includes instructions for setting up the simulation environment and replicating the reported experiments.

5.4 Trust score computation and consensus

Trust scores were computed using the hybrid framework detailed in Sect 3. Direct trust was estimated using an Exponential Moving Average (EMA) model, while indirect trust was calculated using a weighted reputation scheme. These trust values were validated through a hybrid PoS and PBFT consensus mechanism, ensuring fault tolerance and resistance to Sybil attacks. Only transactions endorsed by consensus participants were committed to the blockchain ledger, maintaining consistency and transparency in trust propagation.

5.5 Evaluation metrics and result computation

System performance was evaluated using five primary metrics: (i) trust score accuracy, (ii) malicious vehicle detection rate, (iii) false positive rate, (iv) communication overhead, and (v) trust verification latency. Trust accuracy was validated against ground-truth behavioral classifications. Detection and false positives were analyzed through precision-recall statistics, while communication overhead was measured by counting transmitted trust messages per time slot. Latency encompassed the delay from trust computation to blockchain confirmation.

5.6 Numerical stability and bias mitigation

To ensure the numerical reliability of trust computations, all arithmetic operations were performed using 64-bit floating-point precision. Trust scores were normalized within the [0,1] interval, and adaptive thresholds were used to address rounding errors. Simulations were repeated 100 times using Monte Carlo sampling with varied random seeds to diversify vehicle trajectories and attack patterns. The reported metrics represent the average of these iterations, with 95% confidence intervals. Additionally, k-fold cross-validation was applied to the GaussianNB classifier to avoid overfitting and ensure generalizability.

5.7 Integration with VANET protocols

The proposed decentralized trust framework is designed as an overlay system that operates on top of standard V2V communication protocols, particularly those based on IEEE 802.11p. The trust-related messages are disseminated via the application layer, while core VANET communication protocols remain unmodified. This modular design allows compatibility with existing vehicular communication stacks without requiring modifications to lower-layer protocols. Validator selection in the proposed hybrid consensus mechanism is dynamic and mobility-aware. Validator candidates are selected for each consensus epoch based on a weighted combination of their stake history and recent trust stability scores. This strategy ensures that nodes with transient connections are deprioritized, while consistently reliable nodes maintain validator status. A rotating validator committee approach is adopted to handle topological changes without degrading fault tolerance or consensus reliability.

6.1 Change of indirect trust value of vehicles during active detection

The evaluation of the effectiveness of indirect trust computation through active detection was performed via a simulation involving four vehicle categories with initial indirect trust values set at 0.8, 0.7, 0.6, and 0.2, respectively. Additionally, their accuracy rates for message transmission were established at 90%, 20%, 0%, and 80%, respectively, with active detection conducted every 3 minutes. As illustrated in Fig 5, the indirect trust of vehicles progressively aligned with the accuracy of their transmitted messages as detection occurrences increased. If a vehicle begins disseminating misleading information, its indirect trust experiences a substantial decline. Conversely, a vehicle transmitting accurate data can enhance its trust rating, demonstrating a direct correlation between message accuracy and indirect trust levels. Consequently, our method effectively and flexibly regulates vehicle trust values.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Changes of indirect trust of vehicles during active detection.

https://doi.org/10.1371/journal.pone.0338618.g005

The reported malicious node detection accuracy of 95% originates from the Gaussian Naive Bayes (GNB) classifier that forms part of the trust evaluation layer. The blockchain component does not perform detection itself; instead, it ensures tamper-proof aggregation, propagation, and consensus on trust scores derived from local and indirect interactions. We have revised the text to explicitly separate the contributions of (i) the machine learning layer for accurate behavioral classification and (ii) the blockchain layer for decentralized, secure validation and dissemination of trust values.

The simulation also included changes in indirect trust during active detection under varying weight parameters. The indirect trust alterations of three identical vehicles with weight parameter θ₁ values of 0.1, 0.2, and 0.3, respectively [40], were specifically observed. With a 0% message accuracy rate and an initial indirect trust degree of 0.8, a total of 30 active detection processes were simulated, as depicted in Fig 6. The simulation results indicate that increasing the weight parameter can accelerate the rate of indirect trust changes among vehicles. Therefore, when a vehicle persistently transmits malicious messages, the parameter size can be dynamically modified to enhance the identification speed of malicious nodes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Influence comparison of different weight parameters on the change of vehicle indirect trust in active detection.

https://doi.org/10.1371/journal.pone.0338618.g006

6.2 Trust convergence over time

Fig 7 shows the convergence of trust evaluations over time. The faster a model converges to the correct trust score, the more effective it is in ensuring network reliability.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Trust convergence over time for the proposed hybrid PoS+PBFT model compared to the traditional centralized method.

The proposed model converges faster, reaching accurate trust assessments in fewer iterations.

https://doi.org/10.1371/journal.pone.0338618.g007

As shown in Fig 7, the proposed Hybrid PoS+PBFT model converges significantly faster than the traditional model. It reaches an accurate trust score in less than 60 seconds, whereas the traditional model takes much longer to converge. This fast convergence ensures that the system can quickly and accurately identify trustworthy and malicious nodes, which is critical for the reliability of VANET communications.

6.3 Comparison of consensus algorithms

The PoS-PBFT consensus algorithm proposed in this manuscript is evaluated against several established consensus methods. The Proof of Work (PoW) algorithm offers commendable decentralization and security, featuring straightforward logic and ease of implementation. However, it suffers from excessive resource consumption, prolonged consensus times, and low system throughput due to unnecessary hash calculations [61]. The Proof of Stake (PoS) algorithm enhances resource efficiency by increasing overall throughput and reducing consensus times within the blockchain system. While PoS addresses some of the drawbacks of PoW, it introduces vulnerabilities such as susceptibility to double-spend attacks due to its Coin-Age attribute. Additionally, it tends to favor validators with larger token holdings, which can lead to increased centralization [62]. The Practical Byzantine Fault Tolerance (PBFT) algorithm is recognized for its efficiency and low resource requirements. However, its performance deteriorates significantly when the number of nodes increases, causing both consensus time and resource consumption to rise exponentially. Furthermore, the fault tolerance rate of the Byzantine consensus algorithm is limited to 33%, which is inferior to the 49% fault tolerance rate of proof-based consensus algorithms, making its scalability a concern.

In Fig 8 and Table 5, the hybrid approach achieves lower consensus latency and improved energy efficiency compared to PoW, while maintaining comparable or superior performance relative to PBFT and PoS under increasing network scale. This broader baseline set offers a more balanced and rigorous evaluation of scalability and resource efficiency. Finally, the consensus delay and throughput of these four consensus algorithms are simulated, as shown in Fig 8. Throughput refers to the number of transactions a consensus algorithm can process per second, while consensus latency indicates the time from the submission of new transactions to the final consensus. The results indicate that the PoW algorithm necessitates considerable energy and computational capabilities, leading to reduced throughput (approximately 15 TPS) and prolonged consensus time (around 21.37 seconds). The mining difficulty set for this simulation is relatively low. The PoS algorithm consumes less energy and computing resources, outperforming the PoW algorithm with a consensus delay of 12.67 seconds and a throughput of approximately 70 TPS. The performance of PBFT surpasses that of both PoW and PoS algorithms, featuring lower consensus delay (4.97 seconds) and higher throughput (around 200 TPS). The BFT-DPoS consensus algorithm, while sacrificing some decentralization, demonstrates even more efficient performance than other algorithms, achieving a consensus delay of 3.14 seconds and a throughput of about 1000 TPS.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Performance comparison of different consensus algorithms in terms of (a) Consensus Delay and (b) Throughput.

https://doi.org/10.1371/journal.pone.0338618.g008

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Consensus time for trust validation across different network sizes.

https://doi.org/10.1371/journal.pone.0338618.t005

Also the Fig 8 presents a comparison of consensus delay and throughput across various consensus algorithms, including PoW, PoS, PBFT, BFT-DPoS, and PoS-PBFT. The first graph illustrates the consensus delay, where the proposed PoS-PBFT model achieves the lowest delay of 5 seconds, significantly outperforming other methods. The second graph highlights throughput, showing that PoS-PBFT achieves the highest throughput of 900 transactions per second (TPS), followed by BFT-DPoS with 800 TPS. These results demonstrate the superior efficiency and scalability of PoS-PBFT in blockchain-based VANET applications, making it a robust solution for dynamic Intelligent Transportation Systems (ITS).

With respect to security, PBFT and the Hybrid PoS+PBFT offer the highest levels of protection against Byzantine and Sybil attacks. PBFT’s inherent ability to handle Byzantine faults, combined with PoS decentralized validator selection, ensures robust security. PoW and PoS are also secure but are more vulnerable to specific attack vectors, such as selfish mining in PoW and stake centralization in PoS. The Hybrid model leverages PBFT fault tolerance to further enhance security, making it more resistant to these threats.

Finally, in terms of fault tolerance, PBFT and the Hybrid PoS+PBFT provide the highest degree of resilience. These mechanisms are designed to tolerate Byzantine faults and ensure consensus even in the presence of malicious or faulty nodes. PoW and PoS offer moderate fault tolerance but do not match the robustness of PBFT-based mechanisms. Overall, as shown in Fig 9, the proposed Hybrid PoS+PBFT mechanism consistently outperforms existing consensus algorithms across most performance metrics. By combining the efficiency and scalability of PoS with the low-latency and fault tolerance of PBFT, the Hybrid PoS+PBFT is well-suited for trust management in VANETs, where security, energy efficiency, and real-time performance are critical.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 9. Comparison of consensus mechanisms for VANET trust management.

Performance is evaluated based on latency, energy efficiency, scalability, security, and fault tolerance. The proposed Hybrid PoS+PBFT consistently shows superior performance across these metrics.

https://doi.org/10.1371/journal.pone.0338618.g009

The Hybrid PoS+PBFT mechanism demonstrates its superiority over existing algorithms by providing the best combination of low latency, high energy efficiency, robust scalability, and strong fault tolerance, making it a highly efficient and secure solution for trust management in VANETs.

6.4 Communication overhead vs. time

One of the primary objectives of the proposed system is to reduce communication overhead by selectively propagating trust updates. Fig 10 illustrates the communication overhead over time for both the proposed model and the traditional broadcast-based method.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 10. Comparison of communication overhead between the proposed selective data propagation model and traditional broadcast-based methods over time.

https://doi.org/10.1371/journal.pone.0338618.g010

As shown in Fig 10, the proposed selective data propagation method exhibits significantly lower communication overhead compared to the traditional broadcast-based method. The overhead for the traditional method increases steadily over time, due to the need to broadcast trust updates to all nodes. In contrast, the proposed model reduces overhead by selectively propagating trust information only to relevant nodes, resulting in more efficient network usage, especially as traffic increases. This is particularly beneficial in dense VANET environments where frequent trust updates could otherwise congest the network.

6.5 Trust accuracy vs. number of interactions

The accuracy of trust evaluations is evaluated. Fig 11 presents the trust accuracy, defined as the percentage of correct trust evaluations, as a function of the number of interactions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 11. Comparison of trust accuracy between the proposed decentralized model and a traditional centralized model as the number of interactions increases.

https://doi.org/10.1371/journal.pone.0338618.g011

As shown in Fig 11 shows that the trust accuracy of the proposed model improves with the number of interactions, reaching a high level of accuracy as more data becomes available. The proposed model consistently outperforms the baseline model (which lacks reputational feedback) due to its combination of direct and reputation-based trust assessments. The reputation feedback from neighboring nodes helps the system to make more accurate trust decisions, especially in situations where direct interactions are limited. As a result, the proposed model is more effective in identifying malicious or unreliable nodes in the network.

6.6 Latency vs. number of nodes

Fig 12 demonstrates the latency of trust evaluations as a function of the number of nodes in the network. Latency is a critical performance metric in VANETs, where real-time communication is essential for safety-related applications.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 12. Latency vs. Number of nodes for the proposed model and traditional method.

https://doi.org/10.1371/journal.pone.0338618.g012

As shown in Fig 12, the proposed model maintains relatively low latency even as the number of nodes increases. This is due to the selective data propagation mechanism, which limits the number of trust updates that need to be broadcasted across the network. In contrast, the traditional method exhibits much higher latency as the number of nodes increases, which is a result of the increased communication overhead from broadcasting trust updates to all nodes. The reduced latency in the proposed model is particularly important in scenarios where real-time decision-making is required.

6.7 Comparison of consensus mechanisms

Table 5 compares the consensus time for trust validation between the proposed model and a traditional model across varying numbers of validators. The proposed model consistently demonstrates lower consensus times, with improvements ranging from 30% to 35%, highlighting its efficiency in handling trust validation in blockchain-based VANET environments. This efficiency is particularly significant in dynamic ITS scenarios where timely trust decisions are critical.

A comparative analysis of the performance of different consensus mechanisms is conducted in terms of computational overhead, energy consumption, and latency to further assess the efficiency of the system. As shown in Table 1, the proposed hybrid PoS-PBFT consensus mechanism outperforms traditional PoW and PoS mechanisms in terms of computational overhead, energy consumption, and latency. The hybrid approach leverages the efficiency of PoS for validator selection and the low-latency characteristics of PBFT to achieve fast consensus. This allows the system to operate efficiently even in resource-constrained vehicular environments, ensuring that trust decisions are made quickly without introducing excessive computational load or energy consumption.

The proposed hybrid mechanism achieves lower consensus latency and higher throughput compared to PoW, PoS, PBFT, and BFT-DPoS, as shown in Fig 9. By combining mobility-aware validator selection with adaptive localized PBFT, the protocol remains stable and efficient despite high-speed vehicular movement, while preserving Byzantine fault tolerance guarantees.

6.8 Fault tolerance vs. number of malicious nodes

Fig 13 compares the fault tolerance of different consensus mechanisms by analyzing their performance as the number of malicious nodes increases. Fault tolerance is crucial in dynamic and adversarial environments like VANETs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 13. Fault tolerance comparison between consensus mechanisms as the number of malicious nodes increases.

The Hybrid PoS+PBFT maintains higher trust accuracy even in adversarial environments.

https://doi.org/10.1371/journal.pone.0338618.g013

Fig 13 shows that the proposed Hybrid PoS+PBFT mechanism maintains high trust accuracy even as the number of malicious nodes increases, outperforming traditional PoW, PoS, and PBFT mechanisms. The hybrid model is capable of allowing a higher percentage of malicious nodes while still ensuring reliable trust evaluations, making it more robust in adversarial VANET environments.

6.9 Energy consumption vs. number of validators

Fig 14 presents a comparison of energy consumption as the number of validators increases. This analysis highlights the energy efficiency of the proposed model in contrast with other consensus mechanisms. As shown in Fig 14, the energy consumption of the Hybrid PoS+PBFT mechanism increases at a much slower rate compared to PoS and PoW as the number of validators increases.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 14. Energy consumption comparison between consensus mechanisms as the number of validators increases.

The Hybrid PoS+PBFT mechanism scales better in terms of energy efficiency.

https://doi.org/10.1371/journal.pone.0338618.g014

6.10 Scalability and partition handling in high-density urban environments

To evaluate the scalability of the proposed hybrid PoS–PBFT mechanism under high-density conditions, we conducted experiments by varying the size of the validator pool to emulate thousands of participating vehicles and multiple RSUs acting as validators. Fig 15 shows the relationship between the number of validators and consensus latency for both the hybrid PoS–PBFT mechanism and baseline PBFT. As the validator pool grows, baseline PBFT exhibits steep latency growth, exceeding 500 ms with 100 validators. In contrast, the hybrid PoS–PBFT mechanism maintains sub-linear latency growth, staying below 300 ms with 100 validators. This is achieved through hierarchical validator grouping and adaptive quorum selection, which reduce consensus message complexity without compromising fault tolerance. These results demonstrate that the proposed blockchain design can support the scale of dense urban vehicular networks.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 15. Consensus latency as a function of validator pool size for hybrid PoS–PBFT and baseline PBFT protocols.

The proposed approach exhibits sub-linear latency growth, demonstrating improved scalability in dense vehicular networks.

https://doi.org/10.1371/journal.pone.0338618.g015

In highly mobile environments, temporary network partitions are inevitable. To handle this, the proposed framework enables vehicles to rely on locally cached trust scores during partition intervals, allowing safety-related decisions to continue without blocking on global consensus. Once connectivity is restored, local validator groups asynchronously synchronize with the global ledger through validator signatures and timestamp ordering, ensuring eventual consistency without service interruption.

Finally, for safety-critical VANET applications, response times typically need to be within a few hundred milliseconds to ensure collision avoidance and reliable message authentication. The hybrid PoS–PBFT mechanism achieves consensus within 200–300 ms even with large validator sets, which is well within these operational time budgets. This low-latency consensus, combined with local trust caching during partitions, ensures that decentralized trust management remains responsive and reliable under realistic vehicular dynamics.

We conducted extensive scalability experiments to examine system behavior under large-scale deployments. Fig 16 presents a comparative analysis of two critical blockchain performance metrics, block finality time and validator selection latency across different consensus mechanisms. The block finality time reflects the duration required to achieve irreversible confirmation of a transaction, while validator selection latency measures the time needed to elect or identify the validator set for each consensus round.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 16. Comparison of block finality time, validator selection latency, and relative performance index across different consensus mechanisms.

The results show that the proposed Hybrid PoS–PBFT approach achieves significantly lower block finality time (320 ms) and validator selection latency (100 ms) compared to traditional PoW and PoS methods, while also outperforming baseline PBFT and BFT-DPoS in terms of overall efficiency. This demonstrates the suitability of the hybrid model for latency-sensitive and dynamic VANET environments.

https://doi.org/10.1371/journal.pone.0338618.g016

The results show that the PoW mechanism incurs the highest delay, with block finality exceeding 2 seconds and validator selection around 500 ms. PoS reduces both metrics significantly but still exhibits delays unsuitable for safety-critical VANET applications. PBFT and BFT-DPoS demonstrate much lower finality and selection times, reflecting their suitability for low-latency environments. The proposed Hybrid PoS–PBFT achieves the best overall performance, with block finality around 320 ms and validator selection latency near 100 ms.

In addition to the absolute values, a relative performance index is overlaid as a line plot, highlighting the overall latency trends across mechanisms. This combined visualization illustrates that Hybrid PoS–PBFT consistently delivers lower consensus delays and faster validator selection than baseline approaches, which is essential for real-time trust management in high-mobility vehicular networks.

6.11 Threats to validity

While the simulation-based approach adopted in this study offers a controlled and repeatable environment for evaluating the proposed blockchain-enabled trust management scheme for VANETs, certain validity threats may influence the generalizability and reliability of the results. In this section, we discuss the primary threats to both internal and external validity and describe the mitigation strategies employed.

The Internal Validity refers to the degree to which the observed outcomes can be confidently attributed to the mechanisms and processes embedded within the proposed framework, rather than to extraneous or uncontrolled factors. In our simulations, we attempted to ensure internal consistency by initializing all experiments with multiple random seeds to prevent deterministic biases. Each simulation scenario was repeated ten times under varying vehicular densities and mobility speeds, and the average values were reported to smooth out the influence of stochastic variability. Furthermore, trust evaluations were validated using redundant verification procedures to detect outliers and anomalies in the calculated scores. All algorithmic components, including the PoS+PBFT consensus process and trust propagation routines, were independently tested using modular validation scripts prior to integration into the whole system. Numerical precision and convergence checks were applied where necessary to guard against computational errors. Together, these measures help establish confidence in the integrity and repeatability of the simulation results. External validity refers to the extent to which the results derived from simulations can be generalized to real-world vehicular environments. Although the simulation setup closely mimics practical scenarios in terms of vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication, trust score propagation, and consensus dynamics, certain real-world factors—such as environmental noise, urban signal obstruction, heterogeneous communication hardware, and unpredictable driver behavior—were not fully replicated in the simulation model. The current implementation also assumes that the Trust Authority (TA) and consensus validators operate under ideal network conditions without delays or security breaches, which might not hold in a live deployment. Despite these limitations, the use of parameterizable models and scalable architecture supports future adaptation to more realistic traffic and network conditions. Future work will incorporate co-simulation environments, such as SUMO integrated with NS-3, to account for real traffic flow and communication delays, and ultimately validate the system in testbed or field conditions.