GA based Adaptive Learning Algorithm for IPv4 Packet Classification

Abstract

Packet classification plays a major role to achieve network service like Quality of service (QoS). The purpose of packet classification is to determine the desired action for each incoming packet. It is achieved by machine learning algorithm Artificial Neural Network (ANN) using Back Propagation (BP). BP is widely used method of the ANN but it fails to predict the type of architecture and iterations or epochs used for packet classification problem. It uses trial and error method to choose the value for training parameters and thus leads to a problem of time complexity. So we propose a hybrid approach of Genetic Algorithm called Real Coded Genetic Algorithm (RGA), specifically developed for the task of finding the optimal solution to the hidden nodes and epochs for training feed forward neural network. These optimal values are embedded into ANN-BP for training. In RGA, both crossover and mutation operations can operate directly with floating-point numbers and hence reduce the problem of time complexity in predicting the solutions when compared to existing methods. The proposed method is compared with Back Propagation Algorithm (BPA), Binary Coded Genetic Algorithm (RGA) with ANN and Adaptive Genetic Algorithm (AGA) with NN. The simulation results show that the proposed method is more efficient in terms of accuracy of packet classification.