Comparative analysis of extreme gradient boosting and TabNet models for spatiotemporal prediction of melioidosis...

Melioidosis, an infectious disease caused by Burkholderia pseudomallei, poses significant public health challenges, particularly in regions where specific environmental factors play crucial roles in its spread. However, traditional risk assessment methods for melioidosis do not comprehensively incorporate the diverse environmental factors that influence the distribution of this bacteria. This paper presents a spatiotemporal analysis of melioidosis transmission in Ubon Ratchathani, Thailand, through a comparative evaluation of extreme gradient boosting (XGBoost) and TabNet models. To model the disease distribution over spatiotemporal scales, various environmental datasets were integrated, including land surface temperature, normalized difference vegetation index, normalized difference water index, and rainfall data. The models were trained and validated on data spanning from January 1, 2013, to December 31, 2022, which were obtained from 219 subdistricts. Our comparative analysis of the two models showed that TabNet outperformed XGBoost, particularly in capturing complex interactions between environmental variables and melioidosis cases, and achieved a higher accuracy score (0.950 for TabNet versus 0.892 for XGBoost). While both models performed similarly in terms of the area under the receiver operating characteristic curve, TabNet exhibited marginally more variability. These results underscore the importance of environmental data for refining predictive models that are used for melioidosis surveillance and management.