Machine learning classification models for predicting chronic pain

Abstract

Abstract Chronic pain is a widespread condition that profoundly affects the daily functioning of many people worldwide, characterized by persistent sensory and emotional discomfort associated with real or perceived tissue injury. This study aims to predict chronic pain based on post-traumatic stress disorder (PTSD), alexithymia, anxiety, pain catastrophizing, stress, depression, and demographic variables as correlates. The analysis included data from 234 males and 307 females experiencing chronic pain in Tehran province between 2022 and 2023. The classification results suggested that PTSD and alexithymia were the most significant predictors, followed by anxiety, depression, and pain catastrophizing. Six different machine learning (ML) classification techniques: Naive Bayes, Decision Tree, Gradient Boosting, Light Gradient Boosting Machine, Support Vector Machine, and Stochastic Gradient Boosting (SGB) were applied to examine a dataset detailing various dimensions associated with chronic pain. The findings in the study classified pain severity into three levels, low, medium, and high, based on quantiles and used six ML models to predict these classes. The SGB model outperformed the others, showing higher accuracy and F1 scores, particularly in predicting the medium pain class. SHAP analysis revealed that psychological factors such as alexithymia, anxiety, PTSD, depression, and stress were significant predictors of pain severity, while age and gender had less impact.