A machine learning–based approach to establish predictors of recurrent tuberculosis in the Guelmim region of Morocco

Despite sustained global efforts, tuberculosis (TB) remains a major public health challenge. Recurrent TB, frequently occurring within two years after treatment completion, poses serious concerns for disease control and patient prognosis. Early identification of individuals at high risk of recurrence is therefore critical to improving outcomes and reducing disease transmission. This study aimed to develop a machine learning-based predictive model for recurrent TB in the Guelmim region of Morocco. A random forest-based method was applied for feature selection, followed by the development of seven machine learning models, including logistic regression, random forest, support vector machine, k-nearest neighbours, extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and gradient boosting machine (GBM). Models were trained on 80% of the data using five-fold cross-validation and evaluated on an independent 20% test set, with performance assessed using several metrics, particularly AUROC and PR-AUC. Key predictors of recurrence included weight (12.54%), age (12.07%), and household size (8.57%). Among the seven machine learning models evaluated, XGBoost achieved the highest discriminative performance (AUROC = 63.4%), while LightGBM demonstrated the highest sensitivity (88.6%), and Random Forest yielded the best precision (19%), F1-score (21.6%), and PR-AUC (12.9%).