Enhancing Employee Productivity Prediction through Self-Supervised Learning and Reinforcement Learning

This paper develops an innovative framework that integrates Self-Supervised Learning (SSL) with Reinforcement Learning (RL) to analyze workstation activities for predicting and enhancing employee productivity. The traditional productivity prediction approaches built on supervisor assessments together with performance evaluation yield uncertain and subjective results. The framework relies on unprocessed workstation activity records, which include typing movements combined with cursor movements together with software application usage and user interaction patterns to create meaningful attributes through Self-Supervised Learning. A set of extracted features guides an RL agent during decision-making processes, which leads to increased productivity outcomes. Real-world data evaluation confirmed that the Hybrid SSL + RL model delivers superior performance to SVM, along with SSL-only and RL-only models. Acceptance for the Hybrid model reached 92.7%, while the SSL model achieved 84.3% accuracy, and the RL model scored 90.2% accuracy. Within this study, the Hybrid model showed the maximum precision levels (91.8%) and recall (93.0%), thus indicating its ability to make precise productive predictions. During 40 iterations, the RL agent substantially enhanced its performance until it reached an average reward rate of 0.95, indicating its enhanced ability to improve productivity strategies over time. The Hybrid model displayed adaptability across different tasks at 94.2%, whereas the baseline stood at 85.0%, SSL at 90.5%, and RL at 92.1%. The research proves that integrating SSL with RL results in a productive system that predicts and optimizes workplace employee performance in changing work conditions.