Anomaly detection combining one-class SVMs and particle swarm optimization algorithms

Anomalies are patterns in data that do not conform to a well-defined notion of normal behavior. One-class Support Vector Machines calculate a hyperplane in the feature space to distinguish anomalies, but the false positive rate is always high and parameter selection is a key issue. So, we propose a novel one-class framework for detecting anomalies, which takes the advantages of both boundary movement strategy and the effectiveness of evaluation algorithm on parameters optimization. First, we search the parameters by using a particle swarm optimization algorithm. Each particle suggests a group of parameters, the area under receiver operating characteristic curve is chosen as the fitness of the object function. Second, we improve the original decision function with a boundary movement. After the threshold has been adjusted, the final detection function will bring about a high detection rate with a lower false positive rate. Experimental results on UCI data sets show that the proposed method can achieve better performance than other one class learning schemes.