Predicting employee absenteeism for cost effective interventions

摘要: Abstract This paper describes a decision support system designed for Belgian Human Resource (HR) and Well-Being Service Provider. Their goal is to improve health well-being in the workplace, this end, task identify groups of employees at risk sickness absence who can then be targeted with interventions aiming reduce or prevent absences. To facilitate deployment, we apply range existing machine-learning methods obtain predictions monthly intervals using real HR payroll data that contains no health-related predictors. We model employee as binary classification problem loss asymmetry conceptualise misclassification cost matrix absence. Model performance evaluated cost-based metrics, which have intuitive interpretation. also demonstrate how approached when costs are unknown. The proposed flexible evaluation procedure not restricted specific domain applied address other analytics questions deployed. Our approach considering wider novel absenteeism prediction.