Sick Leave Data Paper.xlsx

This study presents the application of the item response theory (IRT) and classification algorithms to predict long-term sick leave in the unique environment of a public university. Sick leaves of 30 or more consecutive days are predicted based on job satisfaction scores obtained from an instrument developed <i>ad hoc</i> by using IRT and several individual characteristics. The predictive capability of four different machine learning algorithms was investigated, with the Random Forest, without oversampling, demonstrating superior performance in terms of accuracy (0.93) while the Naive Bayes model, when integrated with the ADASYN oversampling technique, emerged as the most effective predictor of sick leave, based on the confusion matrix and 0.80 accuracy score. These findings validate the potential of both models, establishing a promising foundation for further research. Furthermore, this innovative study illustrates the value of machine learning and IRT in identifying the propensity for long-term sick leave, and their potential as practical tools to aid Human Resources management. By enhancing such predictive insights, academic institutions can take proactive measures to manage and mitigate the impact of absenteeism, fostering a more productive and harmonious work environment.