A fused grey wolf and artificial bee colony model for imbalanced data classification problems

Abstract

The issue of imbalanced datasets, i.e., uneven sample distribution among different classes causes training biases and degrades learning algorithm performance. In past, several solutions for data imbalance handling have been proposed but most of them focus on removing the majority class instances, leading to loss of important information. An alternate strategy to mitigate this issue that has been investigated in literature is minority class samples generation. However, generation of quality synthetic samples for minority class remains an open problem. In this study, a fusion of grey wolf optimizer (GWO) with artificial bee colony (ABC) is proposed to generate good representative samples of the minority class. The combination is analysed because GWO has good exploitation abilities, while ABC is good at exploration. The effectiveness of the proposed method is tested on 20 real-world benchmark datasets and on one real-life application, i.e., scam video classification on YouTube using standard assessment indicators. The performance of the proposed method is compared against 18 state-of-the-art data imbalance handling methods using three classification algorithms, i.e., support vector machine (SVM), k-nearest neighbours (KNN) and decision tree (DT). Our experimental results show an improvement in G-mean score on 18 out of 20 datasets with a maximum improvement of 8% for SVM, and on 17 out of 20 datasets with a maximum improvement of 10.7% for KNN and 6.3% for DT respectively. An improvement in AUC score is also seen on 17 out of 20 datasets for SVM and DT with a maximum improvement of 4.5% and 6% respectively, and on 16 out of 20 datasets for KNN with a maximum improvement of 7.7%. These results show that the proposed method is robust.