Chaotic salp swarm optimization using svm for class imbalance problems

Abstract

In most of the real world applications, misclassification cost of minority class samples can be very high. For high dimensional data, it will be a challenging problem as it may increase in overfitting and degradation of performance of the model. Selecting the most discriminate features is popular and recently used to address this problem. To solve class imbalance problems may optimization algorithms have been proposed in the literature. One among them is bio-inspired optimization algorithm. These algorithms are used to optimize the feature or instance selection. In this paper, a new bio-inspired algorithm called Chaotic Salp Swarm Algorithms (CSSA) were used to find the most discriminating features/attributes from the dataset. We employed 10 chaotic maps functions to assess the main parameters of salp movements. The proposed algorithm selects the important features from the dataset and it is mainly comprised of features selection phase, and classification phase. In the former phase, the most important features were selected using CSSA. Finally, the selected features from CSSA were used to train Support Vector Machine (SVM) classifier in the classification phase. Experimental results proved the ability of selecting optimal feature subset using CSSA, with accurate classification performance. Our observation on different data sets using Accuracy, F-measure, G-Mean, AUC and weighted as indicative metric provide better solution.