Application of Back-Propagation Algorithm in Intrusion Detection in Computer Networks

摘要

Computer systems are prone to attacks by incoming malicious packets which are having information against the Request for Comments (RFC) standards. When the fields of the packets have information that is not part of the standards, then the packets are named as intruders. How to detect all possible intrusion packet in addition to other form of intrusion based on behavior of system, is a challenging task even for the leading Operating System (OS) manufacturer. In spite of existing conventional technologies, artificial neural networks have been explored for intrusion detection with little amount of research. In this research, supervised Artificial Neural Network (ANN) trained by the Back-Propagation Algorithm (BPA) has been implemented with varying values of learning factor, . The implemented system will become foolproof if the ANN is trained with all possible intrusion packet types. This study explains a better way of training the ANN for achieving more than 98% of intrusion detection when minimum number of intrusion packets is given during training ANN. Our experimental results show the performance of intrusion packets detection using back propagation algorithm. Thousand packet information of both normal and intrusion have been considered for implementation. The result of Intrusion Detection (ID) is very close to 99%. The topology of the ANN is (41×10×1). The network converged with 550 iterations. However, very huge amount of packets are to be evaluated to know the complete performance of the developed system.