<![CDATA[This study proposes an integrated K-Means and ARIMA framework for electricity consumption anomaly detection at PT PLN UP3 Semarang. K-Means segments customers into low, medium, and high consumption groups, with the optimal K determined using the Elbow method, Silhouette score, and Davies-Bouldin index. ARIMA models predict consumption trends for each cluster, validated using ACF, PACF, and AIC criteria. Anomaly detection employs residual analysis between actual and predicted consumption. Cluster-specific thresholds are dynamically set: μ ± 2σ for normal residual distributions (p ≥ 0.05), Q3 + 1.5 × IQR for non-normal distributions (p < 0.05), and Median ± 3 × MAD for highly skewed or small samples. The framework achieved a Precision of 0.88, Recall of 0.85, and F1-Score of 0.86, outperforming a clustering-only baseline (F1-Score: 0.72). This robust, segment-based approach, validated on real monthly data (January 2012 - February 2024), enhances anomaly sensitivity, interpretability, and operational relevance for proactive energy management. Future work will integrate engineering-based criteria such as power factor and voltage deviation to detect technical anomalies.]]>
Copyrights © 2025
