An approach to WLS-SVM based on QPSO algorithm in anomaly detection

Abstract

Support vector machines (SVM) can overcome the disadvantage of traditional anomaly detection, which need large sample data and have great effect in real-time detection, but has the disadvantage of slow training velocity. Least squares support vector machines (LS-SVM) can overcome the disadvantage of slow training velocity, but makes the solution lose sparsity and robustness. So a weighted LS-SVM (WLS-SVM) based on quantum particle swarm optimization (QPSO) algorithm is presented, which builds a mixed kernel function, adds self adapting weights in LS-SVM, and solves the linear system of equations with QPSO algorithm. It can increase the performance of LS-SVM, accelerate the rate of convergence, save the memory and always get the least square solution. Applied the improved WLS-SVM in anomaly detection, it shows the improved WLS-SVM is superior to LS-SVM and the improved LS-SVM in training speed and the recognition accuracy, and the application effect is notable.