Features Optimization Techniques for Traffic Classifiers

With the continuous development of Internet technology, accurate classification of network traffic becomes more and more important. Statistics-based traffic classification with extremely accuracy and high expansibility has become the mainstream of this domain. However, this method also has some shortcomings, such as, overabundance of statistical features, and insufficient flexibility of feature vector. We propose an optimal feature vector extraction algorithm, which first extracts the optimal feature vector from original feature set before the classifier executes machine learning and classification, so as to achieve the objective of reducing the dimension of feature vector, saving the classifier´s overhead of memory and computation, and improving the classifier´s flexibility. Experimental results show that this algorithm can significantly decrease the dimension of original feature vector, while endowing classifier with more flexibility.