Predictive Workforce Overtime Forecasting in Government: A Machine Learning Architecture for Job-Classification Seasonal Modeling, Leave-Backfill Prediction, and ERP Budget Alert Integration

Abstract

Overtime expenditure is a persistent and volatile component of government personnel budgets, driven by seasonal demand, unplanned leave, and staffing shortfalls that are difficult to anticipate with traditional budgeting methods. This paper presents a machine learning architecture for predictive workforce overtime forecasting in government, combining job-classification-specific seasonal modeling, leave-backfill prediction, and integration with enterprise resource planning (ERP) budget alerts. The architecture decomposes overtime into seasonal, trend, and leave-driven components, trains classification-aware models, and feeds forecasts into ERP systems that raise budget alerts when projected overtime threatens to breach appropriations. We formalize the forecasting model and evaluation metrics, compare candidate algorithms, and illustrate seasonal overtime patterns by job classification alongside actual-versus-predicted performance. Gradient-boosted and recurrent models achieved the lowest forecast error, and the integrated alerting closed the loop between prediction and budgetary control. The architecture equips government workforce and finance managers to anticipate overtime, plan backfill staffing, and prevent budget overruns before they occur.