AI-Based Smart Energy Monitoring System for Detecting Abnormal Electricity Consumption and Potential Electrical Leakage

People who use residential electricity may find themselves facing sudden rises in their electricity bills without knowing exactly why. Although smart meters offer considerably more detailed measurements than the standard monthly metering, the raw interval data alone does not make it clear why there has been a change in household consumption. This study introduces an AI-driven smart energy monitoring system that sets up a baseline specific to each household, compares the incoming readings with the amount that should be consumed, examines any persistent deviations, and produces alerts for consumers. The framework is intended to distinguish between normal changes in demand and persistent abnormal consumption, and to regard possible electrical leakage as a diagnostic possibility that needs to be verified physically rather than as a condition that has already been proven. The method involves data acquisition, preprocessing, extraction of temporal and statistical features, establishment of the baseline, anomaly scoring, event classification, alert generation, and prediction of billing direction. An exploratory survey of 48 residential electricity users was conducted to examine their billing experiences, perception of the transparency of the existing meters, and acceptance of automated alerts for abnormal consumption. The survey found that 75.0% of the participants had had an unexpectedly high bill in the past 12 months, 81.3% said they had received a bill that was much higher than they had expected, and 72.9% expressed a desire for alerts before monthly billing increases. A 3 × 3 Pearson chi-square test was carried out on the perceived usefulness versus the willingness to adopt the proposed system, yielding a chi-square (4) = 9.353 and p = 0.05285; thus, the null hypothesis could not be rejected at the 0.05 level of alpha. The results support the need for consumer-focused energy analytics while leaving the technical effectiveness of the model for further validation based on additional datasets.