A Novel Meta-heuristic Framework for Solving Power Theft Detection Problem: Cheetah Optimization Algorithm

Abstract

In this paper, a two-level stacking technique with feature selection is used to detect power theft. The first level of this technique uses base classifiers such as support vector machine (SVM), naïve Bayes (NB), and AdaBoost selected by evaluating the F-score and diversity criteria. The appropriate features of the base classifiers are selected using a new feature selection algorithm based on the cheetah optimization algorithm (CHOA). This algorithm ensures diversification and intensification in each step of running by adjusting the Attention parameter of the cheetahs. In the second level, a single-layer perceptron (SLP) model is used to obtain the weight of the base classifiers and combine their predictions. The proposed framework is evaluated on the Irish Social Science Data Archive (ISSDA) dataset, and MATLAB R2020b is used for simulation and evaluation. The results of the accuracy, recall, precision, and F-score, specificity, and receiver operating characteristic (ROC) criteria indicated the high efficiency of the proposed framework in detecting power theft.

Keywords

• power theft detection
• cheetah optimization algorithm
• machine learning
• classification
• feature selection