Abstract: Intrusion detection system is a detection mechanism that detects unauthorized, malicious presents in the computer systems. This anomaly-based system learns about the normal users’ behavior and finds the anomalies by matching with this normal behavior. We used a special type of neural network called backpropagation neural network for learning normal users` behavior. The network traffic that only contains information of normal users is presented to the neural network for learning about the normal users` behavior. When the learning is over, the system is ready for general use. We tested the system performance by using a simulated computer network. We divided the training process of neural network in three different approaches. The neural network is trained with huge, not so huge and small amount of data. We tested the detection capability of the system with huge and small amount of data. It is seen from the performance analysis that the system performs well when trained with small amount of data. We have achieved an overall detection rate of 98% for both known and unknown attacks. Moreover, the system can detect 100% normal user. The system works better than other two research works where they have attack detection rate of 24 and 86% respectively.
Moazzam Hossain and Christian Damsgaard Jensen , 2005. Anomaly Based Intrusion Detection System with Artificial Neural Networks . Asian Journal of Information Technology, 4: 715-720.