Anomaly Detection Algorithm Based On Gaussian Model

In recent years,with the increase of data samples and data diversity,anomaly detection problems are getting more and more attention.Anomaly detection is a special classification problem.Dealing with anomaly detection problems,we need to deal with two categories of problems,which are called target classes and anomaly classes,respectively.Real data sets are often constituted by the target class data and there are very few cases of anormaly class,which is called the class imbalance phenomenon,and it is also a special problem of anormaly detection.In different areas,the researchers use different strategies to solve the problem of anomaly detection,this paper proposed an anomaly detection algorithm based on Gaussian model.Firstly,this paper proposed an anormaly detection algorithm based on Gaussian process model(Gp_GPM).The algorithm combined the zero mean priori of Gaussian process regression theory to train the Gaussian process model,and selected the mean and variance of the posterior probability and the posterior probability as the evaluation index of the anomaly detection,and obtained the corresponding evaluation index threshold.The effectiveness of Gp_GPM in the field of the anomaly detection is verified by applying the algorithm to the artificial simulation data sets and the UCI real data sets.Secondly,this paper proposed an anomaly detection algorithm based on Gaussian mixture model(LVBGMMS).The algorithm aimed at disadvantages of EM algorithm and Bayesian framework model selection method,combined an incremental learning method based on component splitting,selected local Bayesian framework for model selection to train Gaussian mixture model,and obtained the corresponding higher and lower threshold as evaluation index.The effectiveness of LVBGMMS in the field of the anomaly detection is verified by applying the algorithm to the artificial simulation data sets and the KDD99 real data set.Finally,we designed and implemented a prototype system of anomaly detection based on Gaussian model.The system applied anomaly detection algorithm based on Gaussian process model(Gp_GPM)and anomaly detection algorithm based on Gaussian mixture model(LVBGMMS)to satisfy the requirement of different users.