A differential evolution-based method for class-imbalanced cost-sensitive learning

In many real-world applications, it is often the case that the class distribution of example is imbalanced and the costs of misclassification are different. In such circumstances, not classification accuracy but misclassification cost minimization is the primary goal leading to the development of the class-imbalanced cost-sensitive learning. Under-sampling is one of the most important methods in dealing with class-imbalance problems, which uses only a subset of the majority class that is imperfect. In fact, many useful majority class examples are easy to be overlooked. To overcome this deficiency, in this paper, we propose a new adaptive method based on differential evolution for class-imbalanced cost-sensitive learning. We simply denote our proposed method by DE. DE explores the potential useful information contained in these majority class examples and creates an optimal example subset with the goal of low misclassification costs. We compare the performance of DE to its state-of-the-art competitors such as random under-sampling (US) and synthetic minority over-sampling technique (SMOTE) in terms of the total misclassification costs of the resulting base classifiers. The experimental results on 22 class-imbalanced data sets show that our proposed DE is notably better than the widely used random under-sampling (US) and the well-known synthetic minority over-sampling technique (SMOTE).