A Method for Feature Selection and Classification of Network Traffic Flow

Abstract

The invention discloses a business flow feature selection and classification method based on a multi-objective self-adaptive evolutionary algorithm. The method first uses the information gain rate to sort the features, filters out some irrelevant features, and achieves the purpose of fast dimensionality reduction. The adaptive evolutionary algorithm searches the feature space, and uses the features with the highest information gain rate as the initial population, and uses the two objective functions of the inconsistency rate and the feature subset dimension as the evaluation function to select the optimal feature subset. Adaptive crossover and mutation maintain population diversity and ensure the convergence ability of the algorithm. At the same time, the present invention uses the designed three-layer KNN classifier model to classify six multimedia service streams: online standard definition live video, web browsing (Baidu), online audio, web browsing (sina), network voice chat, and online standard definition non-live video. Experimental results show that this method has higher classification accuracy than existing methods.

Description

Technical field [0001] The present invention belongs to the field of pattern identification and classification, and in particular, the present invention relates to a network traffic flow feature selection and classification method based on multi-target adaptive evolution algorithm. Background technique [0002] In recent years, with the rapid development of the Internet, accurate and efficient network flow classification is an important basis for network management. The diversity of network multimedia traffic flow types has brought huge challenges to their classification and identification. Traditional circulatory methods mainly include three types: port-based methods, depth package detection methods, and methods based on multimedia flow statistical features. However, with data encryption, the emergence of new applications and the use of dynamic ports, the first two classification methods will no longer apply. Today, most of the researchers focus on include machine learning class...

AI Technical Summary

Problems solved by technology

However, when the feature dimension is high, irrelevant and redundant features will increase the time complexity of multi-objective optimization

For evolutionary algorithms, improper selection of population initialization, crossover and mutation probabilities will reduce the final classification accuracy and algorithm convergence ability

And most of the current multi-objective feature selection algorithms have an objective function that is the accuracy of the classifier, so the convergence speed is slow and the running time is long

Images