Intrusion detection system based on classification

• Main Authors: Gong Shang-fu, Zhao Chun-lan
• Format: Conference Proceeding
• Language: English
• Subjects: classification; Classification algorithms; Feature extraction; Intrusion detection; intrusion detection system; R-SVM; Redundancy; simplification of pratical information; Support vector machines; Testing; Training
• DOI: 10.1109/ICADE.2012.6330103

With the network security issues being more prominent, the safety of system and network resources become more and more important problem. Intrude detecting (ID) has become a top research topic nowadays. Considering the strong generalization ability, high sorting precision and such advantages the support vector machine (SVM) shows in practices involves small sample, high dimension, we will mainly focus on studying and consummating the SVM methods in intrude detecting. ID always generates huge data sets; such raw data sets are incapable of being training due to its large scale and high dimension and redundancy. Intrusion detection system always has the disadvantages such as over-loaded, occupying too much resource, an extension of training and forecasting time... therefore, the simplification of practical information becomes such a necessity. Recursive support vector machine (R-SVM) and Rough set were used for exacting main features of raw data, and many kinds of classification algorithms were used here and it has been tested by KDDCUP1999 date set. The result shows that, the SVM classification based on R-SVM runs excellent, its accuracy is as good as the SVM classification based on the whole features and considerably reduces the training and testing time.