Intrusion Detection Strategies Using Supervised Learning and Semi Supervised Learning

Abstract

Advancements of the technology always happen with the human needs. Based on that the human began to communicate on the wired telephones, and then it spread to computers to communicate and with the time demand the communication networks emerged as huge one and inevitable. Moreover all sorts of communication fields like Telecom, Television, wireless sensor networks, radar have merged into the communication networks for transferring pool of data. So everybody access this huge data making of the mess of traffic and intruders taking this an advantage resort to undesirable activity called as attack. This invasion activity initiates the need for security of the communication network. Different kinds of network merge into the single communication network forming the large networks and the traffic becomes overloaded in the network. Intruders access the infrastructure and information causing turbulence to the network. Traditional intrusion detection system can identify only the known attacks or labeled data traffic. To recognize the unknown attacks or unlabelled traffic, the feature selection of the data is done prior to the Intrusion detection system. In feature selection the features of irrelevant or non -performing are eliminated to improve the quality of the IDS performance. Feature selection of the dataset is done to optimize the classifier activity or stress to control the intruder. In this thesis, the distance based supervised and semi- supervised learning methodologies are proposed. Divergence based feature selection algorithm based on the Kullbuck Leibler divergence metric, known as KLFS (Kullbuck Leibler Feature Selection) is proposed to reduce the feature set and it is implemented. Classification is done using C4.5 classifier and performance parameters like Accuracy, True Positive Rate, False Positive Rate are evaluated and proved to be the best performing compared to existing methods like MMIFS, DMIFS, RPFMI. Algorithm based on distance metric and for final selection of features the K-Means clustering is...