AI Framework Predicts Individual Absenteeism with High Accuracy

Staff absenteeism poses significant operational costs, particularly in high-demand sectors like healthcare and logistics, where reliable individual-level absence prediction is crucial for workforce planning. Existing prediction methods often fall short by merely reproducing past outcomes rather than forecasting future events, and they typically ignore the sequential nature of attendance histories.To address these limitations, researchers developed a novel Time Series Classification (TSC) framework. This framework distinctly separates historical attendance sequences from future absence labels, enabling genuinely proactive predictions. Given the scarcity of public longitudinal attendance data, a reproducible simulated dataset, calibrated to the UCI dataset, was constructed for the study.The framework was evaluated under severe class imbalance, comparing Binary Focal Loss (BFL) and Geometric Mean (G-Mean) loss functions. Both achieved comparable, strong performance, with BFL requiring parameter calibration while G-Mean adapted automatically. Among the deep learning architectures tested—LSTM, CNN, and the hybrid LSTM-FCN—the LSTM-FCN consistently delivered robust precision and specificity. The study found stable performance with specific batch and window sizes, achieving approximately 80% balanced accuracy on test data.