Forecasting Basic Electricity Production Costs at PT PLN (Persero) Using Advanced Models

Indonesia faces growing challenges in managing fossil fuel subsidies, global energy price volatility, and fiscal pressures on its national budget (APBN). These factors require efficient management of the energy sector to maintain affordable electricity tariffs and ensure financial sustainability. Although the Basic Cost of Electricity Production (BPP) plays an essential role in tariff determination, subsidy planning, and operational budgeting, limited research focuses on BPP forecasting. Most studies focus on minimizing numerical errors but often overlook trend directionality and pattern similarity. Furthermore, BPP calculation remains highly manual and involves complex variables that can introduce inconsistencies. This study compares eight forecasting models: ARIMA, ElasticNet, XGBoost, KNN, LightGBM, SGD, Prophet, and Multiple Linear Regression (MLR), applied to historical BPP data. The evaluation uses quantitative metrics, Root Mean Square Error (RMSE) to assess average predictive error, and qualitative metrics, Directional Accuracy (DA), Spearman’s Rank Correlation, and Dynamic Time Warping (DTW) distance to evaluate alignment with actual directional trends and graphical patterns. The results show that ARIMA achieves the lowest RMSE (517.77) but does not capture directional changes effectively. In contrast, XGBoost strikes the best balance between a competitive RMSE (549.45) and strong pattern alignment, making it the most operationally viable model. The findings highlight the need to combine quantitative and qualitative assessments when selecting forecast models. Reliable BPP forecasting will enhance PLN’s financial planning, optimize subsidy allocations, and bolster national energy security through data-driven electricity sector management.