Seminar Title:An Enhanced J48 Classification Algorithm for Anomaly Intrusion Detection System
25 Jul 2017

Presenter: Mohammed Aljundi

Date: Wednesday, July 26, 2017

Time: 2:30PM

Location: PH2104



The use of Internet could be associated with some risks that are linked with an illegal access of the data that are being shared over the internet. Indeed, this could lead to sending or receiving malicious traffic or attacks through the Internet. Therefore, the Intrusion Detection systems (IDSs) have been developed as efficient measures for ensuring security management against anomaly attacks. Moreover, this could help in protecting the computer systems or networks from external attackers and internal users as it monitors computer system and network traffic. The IDS approach works by analyzing the traffic for possible attacks originating from outside the organization and any possible attacks originating from inside the organization. In this thesis, an enhanced algorithm has been developed, which is based on the use of J48 algorithm in order to improve the accuracy of detection and performance for new IDS. Our enhanced algorithm helps in providing an efficient detection of any possible attacks that could threaten the confidentiality of the network. This involves the use of dataset tests through the integration of various approaches such as Naive Bayes, J48, Random Tree and NB Tree. Furthermore, the NSL KDD intrusion dataset is used while running the experiments. The dataset is divided into two datasets: training and testing that is based on data processing. Then, the feature selection accessed through WEKA application has been used to evaluate the efficiency of using these features. The obtained results suggest that the performance of the proposed algorithm without using features selection is more accurate and effective, as compared with the feature selection process.  The implementation of the proposed algorithm guarantees the classification of the dataset with a detection accuracy that reaches 99.88% for all features using 10-fold cross validation test, 90.01% for supplying test set using full test datasets and all feature and 76.23% for supplying test set using test-21 dataset and all feature.