A aplicação de modelos de aprendizado de máquina na previsão de acidentes escorpiônicos na amazônia legal

Abstract

The incidence of scorpionism emerges as a significant public health issue in tropical and subtropical regions, with the Amazon experiencing a notable increase in cases between 2000 and 2017. Holding 13% of the world's scorpion fauna, this region faces challenges due to climate change and indiscriminate soil use, exerting substantial pressure on biodiversity. This study aimed to integrate scorpion occurrence data, bioclimatic information, land use, and epidemiological data to predict scorpionism incidence in northern Brazil. We collected occurrence and bioclimatic data for four species associated with accidents in distinct endemic areas in the Legal Amazon: Tityus obscurus, T. silvestris, T. metuendus, and T. bastosi. Species distribution models (SDMs) were built using MaxEnt software, and niche overlap among species was analyzed. Scorpionism accident data were obtained from the Notifiable Diseases Information System (SINAN) for the period 2015 to 2022. Additionally, land use data from the LUH2 dataset for present and future scenarios were utilized. The predictive model incorporated SINAN incidence estimates, land use data, climatic information, and climatic suitability derived from SDMs as predictors. The model construction employed Machine Learning algorithms, such as XGBoost and Random Forest, and variable importance was assessed using the SHAP (Shapley Additive exPlanations) method. Results indicate significant challenges for scorpion species in the Amazon due to climate change, with substantial losses of climatically suitable areas for some species. While some species showed a reduction in potential distribution, others, like Tityus silvestris, exhibited an increase. The findings not only highlight the challenges faced by scorpion species in the Amazon but also underscore the potential of Machine Learning techniques in analyzing large datasets. The successful application of algorithms such as XGBoost and Random Forest not only enabled the prediction of scorpionism incidence but also emphasizes the crucial role of these methodologies in fields like ecology, epidemiology, and public health.