Anomaly Detection in Electricity Meter Data to Identify Fraud in Enugu DisCo Customers

Electricity meter fraud and tampering imposes major revenue losses on Nigerian distribution companies and contributes to the persistently poor financial performance of the power sector. This study applied anomaly detection methods to electricity consumption records from 28,000 residential customers of Enugu Electricity Distribution Company across Enugu Urban, Agwu, and Oji River zones in South East Nigeria. Monthly consumption records over 36 months were preprocessed to compute consumption ratios, growth rates, and deviation scores from comparable peer groups. Isolation Forest, Local Outlier Factor, and rule-based statistical threshold approaches were compared for identifying potentially fraudulent accounts. Isolation Forest achieved the best performance on a manually validated test set, with precision of 0.84 and recall of 0.79. Approximately 4.2 percent of accounts were classified as high-anomaly, consistent with industry estimates for distribution fraud rates. Anomalous patterns included near-zero consumption for three or more consecutive months followed by sudden high consumption, and systematic under-reading at approximately 60 percent of the expected peer group consumption level. Geographic concentration of anomalous accounts in Agwu zone suggested potential network-level tampering. The study demonstrates the value of data-driven fraud detection in complementing physical meter inspections. Recommendations include deploying the Isolation Forest model as a monthly fraud risk scoring tool, integrating smart prepaid meter rollout in high-anomaly zones, and collaborating with NERC to establish sector-wide anti-fraud data analytics standards.

Keywords: anomaly detection, electricity meter fraud, Enugu DisCo, Isolation Forest, power sector analytics