<![CDATA[Electricity fraud remains a major challenge for PLN (State Electricity Company), causing significant financial losses and reducing service reliability. This study aims to develop an electricity bill fraud detection system at PLN ULP Telda using a hybrid approach that combines Random Forest and Convolutional Neural Network (CNN) algorithms. Random Forest is effective in handling structured tabular data and identifying important features, while CNN excels in recognizing complex patterns in time-series electricity consumption data. By integrating both algorithms, the system is expected to achieve higher accuracy and reliability compared to a single model. The research methodology includes data collection from smart meters, pre-processing, feature extraction, model training, and evaluation using accuracy, precision, recall, and F1-score metrics. Initial results indicate that the hybrid model improves detection performance, reduces false negatives, and strengthens fraud identification. The final results show that the combined system outperforms the single model, with an accuracy of 91.3%, a precision of 89.7%, a recall of 88.4%, and an ROC-AUC of 90.5%. This research contributes to PLN's ability to detect fraud more accurately and efficiently. The contribution of this research is financial efficiency for PLN, fairer electricity tariffs for customers, and a reference for further research in the application of machine learning for energy fraud detection.]]>
