Research Of Twin Support Vector Machines In Intrusion Detection

As one initiative defense technology, the intrusion detection is always a hot research topic in thefield of network security. It has become an indispensable part of modern computer network securitysystem. Support Vector Machines (SVM) is an important research aspect of the intrusion detection,but the intrusion detection based on traditional SVM hasn’t obtained breakthroughs on training speedand detection accuracy, especially for the large dataset. Twin Support Vector Machines (TWSVM)simplifies the complexity and reduces the training time by solving two smaller-scale quadraticprogramming problems which are such as traditional SVM. This research has offered a new thoughtfor the intrusion detection to improve the above problem.The multiclass classification algorithms based on traditional SVM are vulnerable on trainingspeed when dealing with large dataset. To the question, this paper firstly gave an acute analysis ofTWSVM. Then combining TWSVM and the idea of binary multiclass classification, it proposed anew multiclass classification algorithm (Binary Tree Twin Support Vector Machines，BT-TWSVM).To reduce the error accumulation of binary tree SVM (BT-SVM), the algorithm firstly got clusteringcenters through the clustering algorithm, and then compared the distances between them to determinethe separation sequence of classes. Finally, it was applied to network intrusion detection. Theexperiments were carried on KDD Cup99’s dataset. The results show that the algorithm obtains highdetection accuracy and certain advantages on training speed especially for large dataset. The trainingspeed is approximately two times faster than that in traditional BT-SVM. It is valuable for largedataset processing in the field of network intrusion detection.