Assessing the relationship between electricity theft and customer payment habits: A machine-learning approach

Özay E. C., Çakıt E.

UTILITIES POLICY, cilt.95, ss.1-12, 2025 (SCI-Expanded)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 95
  • Basım Tarihi: 2025
  • Doi Numarası: 10.1016/j.jup.2025.101955
  • Dergi Adı: UTILITIES POLICY
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Social Sciences Citation Index (SSCI), Scopus, Academic Search Premier, International Bibliography of Social Sciences, EconLit, Environment Index, INSPEC, PAIS International, Civil Engineering Abstracts
  • Sayfa Sayıları: ss.1-12
  • Gazi Üniversitesi Adresli: Evet

Özet

This study applies machine learning models to historical data to predict the likelihood of debt repayment among customers previously identified as unlawful electricity users. Key predictive factors include past payment behavior, debt levels, demographic characteristics, and regional electricity leakage rates. While existing research on electricity theft predominantly emphasizes detection, repayment prediction remains underexplored. Moreover, although payment behavior analysis is well-established in the financial sector, its application in electricity is limited. This study addresses this gap by demonstrating the effectiveness of Gradient Boosting Machines (GBMs), which outperformed all other models with an accuracy of 90.98 %, an F1 score of 92.20, and an AUC of 0.91. These results highlight GBMs’ capability to manage imbalanced data and capture complex, nonlinear relationships. The findings offer practical value for electricity distribution companies in Türkiye by enabling proactive risk assessment and targeted revenue recovery strategies.