Using Predictive Analytics to Improve Surveillance of Heat-Related Illnesses During Military Training

Heat-related illnesses are important occupational risks in military personnel, especially for soldiers who do not have experience with hot climate regions. Our study focuses on predictive analysis of heat-related illnesses to improve prevention program. The Royal Thai Army (RTA) Medical Department collected data from conscripts during a 10-week military training program in 2013. To build predictive analytic models, we applied various machine learning and deep learning methods, including generalized linear model (GLM), k-nearest neighbors (kNN), random forests (RF), eXtreme gradient boosting (XGB), deep neural networks (DNN), and convolutional neural networks (CNN). We compared accuracy between different models and we found that the accuracies of predictive models’ outcome range from 53% (CNN) to 91% (XGB). Based on the explainability of data analytics outcomes, we choose to use XGB and RF to identify predicted group and create an actionable plan for reducing incidence of heat-related illnesses in new trials.