Table 1.
Overview of studies that developed machine learning models utilizing EDA for point estimations of pain intensity.
Table 2.
The top 22 most informative features extracted from the EDA signal.
Table 3.
The neural network parameters are optimized through a search space hyperparameter tuning process.
Fig 1.
GA consists of selection, crossover, and mutation steps.
Table 4.
GA and loss function (LossL) parameters are optimized using their search spaces through the hyperparameter tuning.
Table 5.
GD and soft loss function (LossS) parameters are optimized through the hyperparameter tuning process within their search spaces.
Fig 2.
Performance comparison of LossS by GD, LossL by GA, and bootstrap methods; LossS outperforms by yielding a narrower PIW across all PICP values.
Fig 3.
(a) higher softening factor (s) value improves the PICP. (b) A higher Lagrangian multiplier (λ) value results in a slightly higher PICP.
Table 6.
The generalized model results demonstrate how MPIW and NMPIW change as PICP varies.
Table 7.
The mean of the upper and lower bounds for each pain level in the generalized model as PICP varies.
Table 8.
Compared to the generalized model results, the PI widths of the personalized model are wider.
Table 9.
The mean upper and lower bounds for each pain level averaged across participants for various PICP values.
Table 10.
The hybrid model results include PICP, MPIW, and NMPIW.
Table 11.
The mean upper and lower bounds for each pain level averaged across clusters for PICP values of 50%, 75%, 85%, and 95%.
Fig 4.
The distribution of pairwise Euclidean distances for Clusters 1, 2, 3, and 4.
Fig 5.
The hybrid approach, which utilizes a clustering technique, outperforms the other models and is considered a viable option for implementation in clinical settings.