Improvement of SOM Classification Algorithm and Application Effect Analysis in Intrusion Detection

Abstract

Self-organizing maps (SOM) algorithm is a kind of tutorless learning method with good self-organization and visualization. It has been widely used and researched. SOM neural networks can classify unknown categories of data without supervision, but the same type of data in the classification result may correspond to different winning neurons. According to the principle of a category corresponding to a winning neuron, the SOM network classification may have more categories than the actual data categories. In order to solve this problem, this paper presents two improved methods: the improved SOM neural network based on system clustering method and the improved SOM neural network based on k-means algorithm. Two improved algorithms are used to cluster a total of 4000 sets of data from five kinds of network intrusion data, and the clustering results are compared with the fuzzy clustering and generalized neural network fuzzy clustering algorithms. The experimental results show that the SOM algorithm based on the system clustering method is better than the SOM algorithm based on k-means algorithm in the intrusion detection.