With the rapid development of the internet technology, more and more security risksappear on internet. The research on Intrusion detection, which can discover each kind ofinvasion behavior rapidly and effectively, and guarantee the system and the networkresources security, becomes hot. This paper is significant in the theory and practicaldomain using theory tools of data mining and machine learning to abnormal detection ofnetwork packets.This paper proposes a new SVM algorithm based on clustering blocks algorithm to reducethe training time of large-scale data, and improve the detection speed of SVM whileguaranteeing the accuracy of classification. Pretreating the original data with HVDMdistance based on heterogeneous data to solve the problem, which the primary data causethe SVM algorithm to be unable to work because of including the character string data.The introduction of an improved K-means algorithm can effectively reduce the choicedependence of start value. The paper proposes Intrusion Detection formation based on theintegration of the improved SVM algorithm on the common intrusion detection framework.Simulation results of SVM algorithm divided by clustering blocks based on KDDCUP1999data improved the training speed and predicting speed with higher detection accuracy,lower omissions rate and lower false alarm rate. Thus, this paper has the reference tomachine learning and invasion detection system domain. |