Neural networks in intrusion detection systems

Security of an information system is its very important property, especially today, when computers are interconnected via internet. Because no system can be absolutely secure, the timely and accurate detection of intrusions is necessary. For this purpose, Intrusion Detection Systems (IDS) were designed. There are two basic models of IDS: misuse IDS and anomaly IDS. Misuse systems detect intrusions by looking for activity that corresponds to the known signatures of intrusions or vulnerabilities. Anomaly systems detect intrusions by searching for an abnormal system activity. Most IDS commercial tools are misuse systems with rule-based expert system structure. However, these techniques are less successful when attack characteristics vary from built-in signatures. Artificial neural networks offer the potential to resolve these problems. As far as anomaly systems are concerned, it is very difficult to build them, because it is difficult to define the normal and abnormal behaviour of a system. Also for building anomaly system, neural networks can be used, because they can learn to discriminate the normal and abnormal behaviour of a system from examples. Therefore, they offer a promising technique for building anomaly systems. This paper presents an overview of the applicability of neural networks in building intrusion systems and discusses advantages and drawbacks of neural network technology.