Pemodelan Prediktif Konsumsi Energi Listrik di Pabrik Baja Berbasis XGBoost untuk Pengelolaan Sumber Daya Berkelanjutan

Abstract

This research aims to develop a predictive model for electrical energy consumption in steel plants based on XGBoost Regressor, with the goal of supporting sustainable resource management. The dataset used was obtained from the UCI Machine Learning Repository with DOI: 10.24432/C52G8C , covering real-time operational data from a steel plant during 2018, including variables such as energy consumption (Usage_kWh), reactive power, power factor, and production load status. The research process included data exploration, preprocessing, time and categorical feature extraction, and training the XGBoost Regressor model with hyperparameter optimization using Grid Search and time-series split validation. Evaluation results showed outstanding performance with an MAE of only 0.41 kWh, RMSE of 0.81 kWh, and an R² value of 0.9993, indicating that the model successfully explained nearly all variations in actual data. Feature importance analysis revealed that Lagging_Current_Reactive.Power_kVarh and CO2(tCO2) were the most influential features in predicting energy consumption. This model is not only technically accurate but also holds significant practical potential for use in industrial energy management systems, helping companies plan production schedules, avoid peak loads, and improve energy efficiency sustainably.