Network traffic classification using feature selection and parameter optimization

Network traffic classification is the foundation of many network research works. In recent years, the research on traffic classification and identification based on machine learning method is a new research direction. Support Vector Machine (SVM) is the one of the machine learning method which performs good accuracy and stability. However, the traditional classification performance of SVM is not ideal. We proposed an optimized method which can improve the performance of SVM greatly. we extracted feature subset with wrapper approach and calculated the optimal working parameters automatically based on grid search algorithm. We applied this method to two-class SVM classifier. The simulation results validated that all of the flows’ average accuracy reaches 99.64%, average feature dimension reduces 20% than original dimension and average elapsed time is shorter 98.88% than traditional SVM. The optimized method can reduce feature dimension, shorten elapsed time, improve the performance of SVM classifier obviously. © 2015 Journal of Communications.