Sparseness of least squares support vector machines based on active learning

To solve the sparseness problem of least squares support vector machine(LSSVM)learning process, this paper proposes a learning algorithm of LSSVM data sparseness based on active learning. This algorithm first selects initial samples based on a kernel clustering method and constructs a minimum classification using LSSVM, calculates the sample distribution under the action of the classifier, and labels the samples closest to hyper planes. These labeled samples are finally added into the training sets to train a new classifier, and the processes are repeated until the model accuracy meets requirements. The LSSVM sparse model of some samples are established. Experiments on the University of California Irvine(UCI)data sets show that the proposed algorithm can increase the sparseness of LSSVM by more 46 percent and reduce the cost of labeling samples.