A novel multi-classifier based on a density-dependent quantized binary tree LSSVM and the logistic global whale optimization algorithm

The least squares support vector machine (LSSVM) is a useful binary classifier, but its performance is limited due to the lack of sparseness. The density-dependent quantized LSSVM (DSM) with quantized input data can increase the sparseness to effectively accomplish binary classification. However, the DSM cannot be directly used in multi-classification applications for most practical data-classification problems. We propose a novel multi-classifier based on a density-dependent quantized binary tree LSSVM (DBSM) and the logistic global whale optimization algorithm (LWA) to improve multi-classification accuracy and computational efficiency. The DBSM consists of multiple DSM classifiers, which hierarchically divide data according to a modified binary tree architecture. The tree architecture is constructed quickly and correctly with the quantized data instead of the original input data. An appropriate initial population of DBSM parameters is generated by using a logistic map and an improved opposition-based learning strategy. Then, the DBSM parameters are optimized by the whale optimization algorithm integrated with the gbest-guided artificial bee colony algorithm. According to the experimental results, the DBSM solves multi-classification problems faster than the one-versus-one based support vector machine (OVO-SVM) and the one-versus-all based LSSVM without sacrificing accuracy. The LWA precisely finds the optimal DBSM parameters without a heavy computational burden, in contrast to recent optimization algorithms. The proposed classifier achieves a 3.39% higher accuracy and consumes 52.83% less time than the genetic algorithm-based OVO-SVM. These results prove that the LWA-DBSM can complete multi-class classification tasks precisely and quickly.