Effective Use of Data Mining in Intrusion Detection

Abstract

ABSTRACT newlineComputer systems connected through network plays an important role in day to day life. Data transmission through network has increased from kilo bytes to terra bytes. People in large enterprises transmit data for their business dealings, People at academic side transmit data to share knowledge, experts at various departments share their conclusions etc. All these will be targeted by the hackers, who disturb the normal transmission of data. Hence an effective mechanism to identify such a misbehaviour activity is needed. There are intrusion prevention systems and various authentication methods to tackle this problem. All these mechanisms face various drawbacks in identifying the hackers. With these challenges, Intrusion detection system (IDS) evolved, where the system analyzes the audit data for intrusion. The performance of the IDS can be improved by better analysis of the audit data. The analysis task can be better handled by data mining algorithms. The main aim of IDS is to have reliable detection and cope with large amount of network traffic. newlineThis research aims to develop a novel hybrid systematic framework to semi-automate the process of building intrusion detection systems. A basic premise is that when audit mechanisms are enabled to record system events, distinct evidence of legitimate and intrusive activities will be manifested in the audit data. In this research work intrusion detection is considered as a data analysis task. The objective is to design a high performance IDS with the help of data mining algorithms, which improves the accuracy and efficiency of the overall IDS and reduces False positive rate and False negative rate. newlineIDS effectiveness can be improved by developing an accurate detector module. The accuracy of the detector module is decided by the detecting patterns stored in the knowledge base. Detecting patterns can be generated either by supervised learning and unsupervised learning method. Unsupervised Learning methods learn with instant pattern but it may result in high False Nega