Hierarchical Attention Models for Predicting Detailed Employee Performance Trends

Predicting employee performance with fine-grained temporal resolution is vital for effective workforce management. This study introduces Temporal Hierarchical Attention Networks (THAN), a novel approach for forecasting employee performance over time. THAN employs hierarchical attention mechanisms to capture long-term and short-term temporal dependencies. It utilizes a hierarchical structure to capture patterns across various time intervals and a temporal attention mechanism to focus on relevant performance features within each interval. Experimental results on real-world datasets demonstrate THAN's effectiveness compared to baseline methods like ARIMA and Long Short-Term Memory (LSTM) networks. THAN achieves an average improvement of 15% in performance forecasting accuracy over ARIMA and 10% over LSTM. These findings highlight THAN's capability to capture fine-grained temporal dynamics, offering a promising framework for enhancing workforce planning and decision-making in organizations.