PREDICTION OF ANOMALY IN HOUSEHOLD ELECTRIC CONSUMPTION USING RANDOM FOREST AND PROPHET

ABSTRACT The development of intelligent systems is essential for improving decision-making in dynamic environments, particularly in identifying abnormal electricity consumption in households caused by energy theft, faulty appliances, or incorrect meter readings. This study combines two machine learning models: Random Forest and Prophet. Random Forest models the relationship between electricity usage and influencing factors, while Prophet forecasts expected consumption patterns based on time-related trends such as daily and seasonal variations. Smart meter data was collected, cleaned, and processed. The models were trained and evaluated against actual electricity consumption. Any significant deviation from expected usage was flagged as an anomaly. Results showed that Random Forest outperformed Prophet and ARIMA, achieving: Mean Absolute Error (MAE): 0.28 kW Root Mean Squared Error (RMSE): 0.41 kW The study concludes that combining Random Forest and Prophet provides an effective and practical solution for anomaly detection in electricity consumption, aiding in energy management, fault detection, and waste reduction.