Hybrid Machine Learning Framework for Electricity Theft Detection and Consumption Anomaly Analysis in Nagpur Distribution Networks

Abstract

Electricity theft and electricity consumption anomalies are two of the biggest problems faced by today\'s distribution network systems, leading to huge losses and inefficiencies. Traditional methods of theft detection are mainly based on manual inspection and are costly and ineffective in detecting more sophisticated consumption patterns. With the emergence of smart meters, huge amount of electricity consumption data can now be analyzed with machine learning methods to detect suspicious patterns. This paper aims to design a Hybrid Machine Learning Framework for Detection of Electricity Theft and Consumption Anomaly Analysis in Nagpur Distribution Networks. The proposed system will use hybrid methods of supervised and unsupervised learning algorithms to detect both types of known and unknown electricity theft patterns. Deviation analysis using a rule-based method is first used to detect any significant deviation in the consumption pattern.

Introduction

The text presents a Hybrid Machine Learning Framework for Electricity Theft Detection and Consumption Anomaly Analysis in Nagpur Distribution Networks. Electricity theft is a major problem for power distribution companies because it causes financial losses and reduces the reliability and efficiency of the electricity distribution system.

With the increasing use of smart meters, large amounts of electricity consumption data are available. This data can be analyzed using Machine Learning (ML) to automatically identify unusual consumption patterns. Traditional theft-detection methods depend on manual inspection and fixed rules, making them costly, slow, and difficult to apply to large-scale distribution networks.

The proposed approach combines several techniques because a single ML model may not detect every type of theft or abnormal consumption. The framework uses:

• Random Forest (RF) and Histogram Gradient Boosting (HGB) for supervised detection of known electricity-theft patterns.
• Isolation Forest (IF) for detecting global or unusual consumption patterns.
• Local Outlier Factor (LOF) for identifying anomalies based on the behavior of neighboring data points.
• Rule-Based Deviation Analysis to compare current consumption with expected or normal behavior.
• Ensemble Logic to combine the outputs of these methods and improve detection reliability while reducing false positives.

Previous research has demonstrated that machine learning, deep learning, anomaly detection, and hybrid approaches can effectively identify electricity theft. Studies using techniques such as CNN, GRU, LSTM, LightGBM, k-means SMOTE, and Transformer-based models have shown potential for handling complex, temporal, and previously unseen theft patterns.

However, the proposed framework has several limitations. These include lack of real-world testing, poor or incomplete smart-meter data, class imbalance, high computational cost, offline operation, and false positives caused by legitimate changes in consumer behavior or seasonal consumption.

The findings suggest that hybrid and ensemble models are more robust than individual models, particularly because supervised methods can identify known theft patterns while unsupervised methods can detect previously unknown anomalies. Deep learning and periodicity-based techniques can further improve the understanding of complex consumption behavior.

Future improvements include integrating LSTM, CNN, and Transformer models, real-time smart-meter data streaming, IoT and cloud technologies, Explainable AI (XAI), continuous model retraining, real-world validation, and mobile/web-based alert systems for electricity officials.

Conclusion

The presented Hybrid Machine Learning Framework for Electricity Theft Detection and Consumption Anomaly Detection is aimed at automatic detection of any suspicious consumption of electricity. Rule-based deviation analysis, supervised and unsupervised learning methods are combined in order to find not only known electricity theft cases, but also new consumption anomalies. Known electricity theft cases are recognized with the help of Random Forest and Histogram Gradient Boosting classifiers, while Isolation Forest and Local Outlier Factor (LOF) find global and local anomalies. It is supposed that the combination of these methods will increase the effectiveness of the detection process and will decrease the probability of false alarms. Furthermore, the developed system can help electricity companies with detecting any abnormalities in consumption via the dashboard. In such a way, there will be no need for manual detection of the anomalies.

References

[1] Y. Zhang, L. Wang, and H. Liu, “A Machine Learning-Based Detection Framework Against Intermittent Electricity Theft Attacks,” International Journal of Electrical Power & Energy Systems, vol. 149, Jul. 2023. [2] S. Kumar, R. Singh, and A. Verma, “An IoT-Enabled Machine Learning Model for Energy Theft Prevention System,” Journal of Cloud Computing, vol. 12, no. 1, 2023. [3] R. Mehmood and K. Abbas, “Systematic Review of Household Electricity Anomaly Detection Using Machine Learning,” Journal of Electrical Systems and Information Technology, 2023. [4] M. Ahmed, A. Khan, M. Ahmed, M. Tahir, G. Jeon, G. Fortino, and F. Piccialli, “Energy theft detection in smart grids: Taxonomy, comparative analysis, challenges, and future research directions,” IEEE/CAA Journal of Automatica Sinica, vol. 9, no. 4, pp. 578–600, Apr. 2022, doi: 10.1109/JAS.2022.105404. [5] Y. Sun, X. Sun, T. Hu, and L. Zhu, “Smart grid theft detection based on hybrid multi-time scale neural network,” Applied Sciences, vol. 13, no. 9, p. 5710, 2023, doi: 10.3390/app13095710. [6] K. Sim, G. Lee, and Y. J. Shin, “Detection and location of electricity theft via convolutional neural network in distribution system,” in Proc. 2023 IEEE Power & Energy Society General Meeting (PESGM), 2023, doi: 10.1109/PESGM52003.2023.10253338. [7] L. Grant, H. Latchman, and K. Alli, “Detection of electricity theft in developing countries—A machine learning approach,” Trends in Computer Science and Information Technology, vol. 8, no. 2, pp. 038–049, 2023, doi: 10.17352/tcsit.000067. [8] H. Fang, J.-W. Xiao, and Y.-W. Wang, “A machine learning-based detection framework against intermittent electricity theft attack,” International Journal of Electrical Power & Energy Systems, vol. 150, p. 109075, 2023, doi: 10.1016/j.ijepes.2023.109075. [9] H. Iftikhar, N. Khan, M. A. Raza, et al., “Electricity theft detection in smart grid using machine learning,” Frontiers in Energy Research, vol. 12, 2024, doi: 10.3389/fenrg.2024.1383090. [10] N. Khan, M. A. Raza, D. Ara, et al., “A novel deep learning technique to detect electricity theft in smart grids using AlexNet,” IET Renewable Power Generation, 2024, doi: 10.1049/rpg2.12846. [11] M. Z. Gündüz and R. Das, “Smart grid security: An effective hybrid CNN-based approach for detecting energy theft using consumption patterns,” Sensors, vol. 24, no. 4, p. 1148, 2024, doi: 10.3390/s24041148. [12] M. T. Quasim, K. u. Nisa, M. Z. Khan, M. S. Husain, S. Alam, M. Shuaib, M. Meraj, and M. Abdullah, “An internet of things enabled machine learning model for Energy Theft Prevention System (ETPS) in Smart Cities,” Journal of Cloud Computing, vol. 12, art. no. 158, 2023, doi: 10.1186/s13677-023-00525-4. [13] S. O. Tehrani, A. Shahrestani, and M. H. Yaghmaee, “Online electricity theft detection framework for large-scale smart grid data,” Electric Power Systems Research, vol. 208, p. 107895, 2022, doi: 10.1016/j.epsr.2022.107895. [14] F. G. K. Guarda, B. K. Hammerschmitt, M. B. Capeletti, N. K. Neto, L. L. C. dos Santos, L. R. Prade, and A. Abaide, “Non-hardware-based non-technical losses detection methods: A review,” Energies, vol. 16, no. 4, p. 2054, 2023, doi: 10.3390/en16042054. [15] Y. Chen, J. Li, Q. Huang, K. Li, Z. Zhao, and X. Ren, “Non-technical losses detection with Gramian angular field and deep residual network,” Energy Reports, vol. 9, pp. 1392–1401, 2023, doi: 10.1016/j.egyr.2023.05.183. [16] J. Huang and Y. Qiu, “LSTM-based time series detection of abnormal electricity usage in smart meters,” Preprints.org, 2025, doi/Manuscript: 202506.1404. [17] [15] A. J. Almalki, “Unsupervised Learning With Hybrid Models for Detecting Electricity Theft in Smart Grids,” IEEE Access, vol. 12, pp. 187027–187040, 2024, doi: 10.1109/ACCESS.2024.3498733. [18] C.-W. Tsai, “An effective ensemble electricity theft detection algorithm for smart grid,” IET Networks, 2024, doi: 10.1049/ntw2.12132. [19] N. M. Elshennawy, D. M. Ibrahim, and A. M. Gab Allah, “An efficient electricity theft detection based on deep learning,” Scientific Reports, vol. 15, Art. no. 12866, 2025. [20] Z. Wu and Y. Wang, “PE-DOCC: A Novel Periodicity-Enhanced Deep One-Class Classification Framework for Electricity Theft Detection,” Applied Sciences, vol. 15, no. 4, p. 2193, 2025, doi: 10.3390/app15042193. [21] T.-D. Le, N. T. M. Duy, T. H. B. Huy, P. V. Phu, H. T. Doan, and D. Kim, “Advanced deep learning-based electricity theft detection in smart grids using multi-dimensional analysis with Convolutional Autoencoder and Transformer,” Engineering Applications of Artificial Intelligence, vol. 157, Art. no. 111333, 2025, doi: 10.1016/j.engappai.2025.111333. [22] “Detecting electricity theft in smart grids using optimized machine learning ensemble techniques and eXplainable AI,” Energy Reports, vol. 14, pp. 3142–3162, 2025, doi: 10.1016/j.egyr.2025.09.019.

Copyright

Copyright © 2026 Prof. Ashish Nanotkar, Anjali Lakdeswar, Devashree Shamkule, Babi Urade, Mansi Awachat. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.