A New Algorithm Combining Self Organizing Map with Simulated Annealing Used in Intrusion Detection

The effect of clustering by Self Organizing Map (SOM) is always effective in intrusion detection (IDS). But there are still some limitations in the algorithm of SOM, such as the algorithm is easy to get into the local minimum, detection accuracy is low, the convergence speed is slow and so on. In this paper, to improve the accuracy and convergence rate, we use Simulated Annealing (SA) algorithm to refine the weight of SOM. SA algorithm find the optimal point by a form of probability, and it is proved that if enough time is given, the SA can certainly find the optimal point. The algorithm is divided into two steps: first, use traditional SOM algorithm to train samples; second, adjust the weight of excited neuron and its neighborhoods by SA algorithm. The simulation experiment results illuminate that the application performs fairly more effective.