Utilizing Stacking Ensemble Algorithm for Employee Productivity Prediction (Case Study of PT PLN (Persero) Unit Induk Distribusi Sulawesi Selatan, Sulawesi Tenggara dan Sulawesi Barat)

Employee productivity is a critical determinant of organizational efficiency and value, particularly in state-owned enterprises such as PT PLN (Persero) in Indonesia’s electricity sector. Accurate prediction of employee productivity can enhance human resource management and support strategic decision-making. This study aims to develop a predictive model for employee productivity using machine learning algorithms based on employee identity, attendance trends, certifications, training participation, and performance targets measured via the NSK indicator. Nine machine learning algorithms were implemented, including Linear Regression, Random Forest, Decision Tree, XGBoost, MLP, SVR, CatBoost, LightGBM, and ExtraTrees, with data balancing performed using the SMOGN technique to address imbalanced target distributions. Model performance was evaluated with MAE, MSE, RMSE, R², and sMAPE metrics. Initial results revealed suboptimal predictions, prompting the application of a stacking ensemble approach combining the three best-performing base models. The optimized stacking model achieved an RMSE of 11.24, MAE of 3.18, R² of 0.58, and sMAPE of 1.03, with residual analysis confirming improved prediction accuracy. Among individual models, CatBoost Regression performed best, achieving an MAE of 0.8. These findings indicate that machine learning, particularly CatBoost, can effectively predict employee productivity and provide actionable insights for HR management. The proposed model supports data-driven decision-making, enabling PLN to optimize workforce allocation, monitor performance targets, and develop strategic initiatives to improve organizational efficiency.