Skip to main content
Advertisement

< Back to Article

Fig 1.

Monthly dengue cases in Kanchanaburi Province, Thailand, from 2014 to 2023.

The time series displays substantial temporal variability and recurrent outbreak peaks, supporting the application of dynamic time-series modeling approaches.

More »

Fig 1 Expand

Table 1.

Summary statistics for monthly dengue incidence and climatic variables in Kanchanaburi Province.

More »

Table 1 Expand

Table 2.

Spearman’s rank correlation coefficients between monthly dengue cases and climatic variables from two meteorological stations in Kanchanaburi Province.

More »

Table 2 Expand

Table 3.

Posterior summaries and convergence diagnostics of the Bayesian negative binomial model.

More »

Table 3 Expand

Table 4.

Incidence Rate Ratios (IRR) and posterior diagnostics from the Bayesian negative binomial model for dengue incidence in Kanchanaburi.

More »

Table 4 Expand

Fig 2.

Posterior predictive check (PPC) density overlays for monthly dengue incidence.

Panels (A) and (B) correspond to the two study areas. The observed dengue cases are compared with 100 replicated datasets drawn from the posterior predictive distribution of the Bayesian negative binomial model. The close agreement between observed and replicated densities indicates that the model adequately captures key marginal distributional features, including overdispersion and right-skewness.

More »

Fig 2 Expand

Fig 3.

Bayesian posterior predictive autocorrelation function (ACF) checks for monthly dengue incidence.

Panels (A) and (B) correspond to the two study areas. The solid blue line represents the posterior median ACF computed from replicated datasets drawn from the posterior predictive distribution; the shaded band denotes the 90% posterior predictive envelope, and the red points indicate the observed ACF. Autocorrelations are displayed up to lag 40 months to assess whether the fitted model reproduces the short- to medium-term temporal dependence observed in the data.

More »

Fig 3 Expand

Fig 4.

Autocorrelation function (ACF) of Bayesian-consistent Pearson residuals from the negative binomial regression model with lagged dengue cases and climate covariates.

Panels (A) and (B) correspond to the two study areas. The dashed lines denote approximate 95% confidence bounds under the white-noise assumption. Residual autocorrelation is examined up to 60-month lags to evaluate potential long-term temporal structure, including the documented 3-5 year dengue outbreak cycles. The absence of substantial residual autocorrelation across lags indicates that temporal dependence is adequately captured by the lagged response structure without the need for an additional stochastic autoregressive error component.

More »

Fig 4 Expand

Table 5.

Model comparison based on leave-one-out cross-validation (LOO) and predictive accuracy.

More »

Table 5 Expand

Fig 5.

Out-of-sample forecasting performance across k-step-ahead horizons (1-6 months).

Panels (A) and (B) report RMSE and MAE, respectively, comparing the predictive accuracy of Bayesian negative binomial (teal) and Poisson (orange) models. Climate covariates from the Kanchanaburi and Thong Pha Phum meteorological stations are used to assess model stability and spatial representativeness. Both metrics indicate that the negative binomial model consistently outperforms the Poisson model across all forecast horizons and study areas.

More »

Fig 5 Expand