Analyzing Unrest Events in Thailand's Southern Provinces From 2004 to 2020 Using Machine Learning Classification Models

ABSTRACT The ongoing unrest in Thailand's southern provinces; Narathiwat, Pattani, Yala, and parts of Songk hla poses significant challenges due to its historical, cultural, and socio‐political complexities. This study combines traditional machine learning classifiers (Naive Bayes, SVM, and Decision Tree) with ensemble learning techniques (Random Forest). These classification models are used to predict the multiclass of unrest victimization based on data from the Deep South Coordination Center (DSCC) spanning 2004 to 2020. The outcome variable comprises three categories: safe, injured, and dead, forming a multiclass classification task. Data preprocessing and feature selection were performed using the chi‐squared test, and model implementation was carried out using R libraries e1071 and randomForest . Among the models, Decision Tree achieved the highest average F1 score (74%), followed by NB (71%), SVM (67%), and RF (64%). DT also demonstrated strong performance in classifying the “safe” and “injured” categories. Key predictive features include the type of attack, geographical location, and event timing. These results support the integration of such models into real‐time decision support systems to enhance conflict monitoring and resource allocation. Limitations include reliance on a single dataset; future work should explore additional data sources and advanced models such as Convolutional Neural Networks (CNNs) to further improve predictive accuracy.