Abstract
The rising concern over social worker turnover underscores the need to identify its determinants in order to promote occupational stability. In recent years, machine learning techniques have emerged as powerful tools for predicting individual risk behaviors and uncovering the underlying drivers of workforce attrition. This study applied and compared three widely used machine learning algorithms, namely Random Forest, Light Gradient Boosting Machine (LightGBM), and Support Vector Machine, to identify key predictors of turnover intention among social workers. Predictors were analyzed across individual, organizational, and value-based dimensions. Among the models tested, LightGBM achieved the best overall performance. To enhance interpretability, Shapley Additive Explanations were employed to quantify and visualize the contribution of each predictor. The eight most influential predictors identified were depersonalization, organizational commitment, job satisfaction, social belief, salary satisfaction, managerial support, marital status, and age. These findings demonstrate the utility of machine learning approaches for forecasting turnover intention and provide actionable evidence to inform the development of targeted retention policies and strategies.