Technical losses in the electric power distribution system are a key indicator of the efficiency of energy delivery from the provider to consumers. This study aims to analyze and predict technical losses in the UP3 Banten Utara power distribution system using the Automatic Relevance Determination (ARD) Bayesian Regression method based on
machine learning. The dataset used consists of actual technical loss percentages from January 2023 to December 2024. The research stages include initial trend analysis, generation of derived features (lag, moving average, percentage change), data normalization using the Z-score method, training and testing data splitting, model training, and evaluation using error metrics such as Mean Absolute Percentage Error (MAPE) and residual analysis. The results show that the ARD Bayesian Regression model is capable of closely following historical patterns, with predicted values showing high similarity to actual data. Long-term predictions for 2025–2027 indicate relatively stable
seasonal fluctuations ranging from 2.88% to 3.57%. These findings can serve as a reference for PLN in planning strategies to control technical losses, particularly during predicted peak periods, to minimize energy losses and improve distribution efficiency