Electricity Theft Detection Employing Machine Learning Algorithms

Abstract

Electricity theft is one of the biggest problems for smart grids. As a result of the reliance of current procedures on certain equipment, it cannot be easily detected. Additionally, the techniques don't effectively extract useful information from highly dimensional power usage data, which raises the incidence of false positives and restricts their output. This paper intend to design and develop a machine learning classifier that can distinguish between legitimate customers and those who are committing power theft by tampering with electricity meters or by other means. Prepaid users will be the ones we test the most for this, along with their long-term monthly bill history. In this paper, we tried to classify electricity users based on their regular bills and any notable changes to those bills, and then further categorise them into customers who are genuine and those who are responsible for fraud and theft. We have applied several machine learning techniques: Logistic Regression (LR), Support Vector Machines (SVM), Naïve Bayes Classifier (NB), Decision Tree (DT), Random Forest (RF) and Adaptive Boosting (AdaBoost) in order to discover the power theft and stop additional losses to the grid. The main motivation of this work is to find the best machine learning model that is the most effective, conserve electricity from waste, and prevent economic loss at the same time.