An Explainable Risk-Based Smart Meter Analytics Framework for Non-Technical Loss Assessment and Mitigation in Electrical Distribution Networks

Abstract

Non-technical losses (NTLs) continue to subvert revenue recovery, energy accounting, and operational efficiency in electrical distribution networks. In this paper, we propose an explainable risk-based smart meter analytics framework, which includes behavioral feature engineering, consumer profiling, hybrid anomaly detection, model interpretation, energy loss estimation, and mitigation prioritization. Temporal and consumption-related descriptors were extracted from smart meter measurements and consumers with heterogeneous load characteristics were clustered using K-means. For unsupervised detection of novel anomalous consumption patterns, Isolation Forest was applied and for supervised classification of NTL, XGBoost was utilized. SHapley Additive exPlanations (SHAP) were used to offer global and consumer-level explanations of the model predictions. Subsequently, a composite risk score comprising the predicted NTL probability, consumption deviation, and estimated energy loss, was developed to prioritize inspection and mitigation actions. The framework was tested on 175,200 observations from 500 consumers in the years 2020-2024. The proportion of outliers was 10%. We identified three consumer profiles with a silhouette coefficient of 0.487 and a Davies–Bouldin index of 0.71. The hybrid Isolation Forest–XGBoost framework achieved higher performance compared to the evaluated standalone classifiers, with an accuracy of 96.8%, precision of 0.95, recall of 0.94, F1-score of 0.94, and receiver operating characteristic area under the curve of 0.98. Consumer level risk assessment categorized the consumers into high risk (42, 8.4%), medium risk (96, 19.2%) and low risk (362, 72.4%). These results demonstrate that the proposed framework can accurately detect NTL with transparent explanations and impact-based prioritization, thereby providing an operational decision-support tool for targeted inspections and loss-mitigation interventions.