Abstract: A big data-based model is proposed for diagnosing electricity usage anomalies. By extracting key features from electricity, load, and event data, and applying feature selection, 38 core indicators are retained. Ensemble tree algorithms handle data imbalance and nonlinearity effectively. Experiments show superior performance over traditional methods in accuracy, recall, and F1-score. The model supports intelligent anomaly detection and offers strong potential for real-world deployment in electricity inspection systems.