A discriminative dictionary learning-AdaBoost-SVM classification method on imbalanced datasets

Abstract

Sparse representation is a signal processing method which is mostly used in signal compression, noise reduction, and signal and image restoration fields. In this study, sparse representation was used in a different way from the traditional methods. In the proposed method, a hybrid structure was created by combining dictionary learning and ensemble classifier AdaBoost algorithms. The main idea of this method is to obtain the sparse coefficients from an over-complete dictionary and to use the coefficients in the weight update formula of AdaBoost. Support Vector Machines (SVM) classifier was used as weak classifiers of AdaBoost, and AdaBoost-SVM classifier structure was created. Multiplying the sparse coefficients with weight of weak learners process in weight update formula has given satisfying results on imbalanced datasets during the experiments.