A Feature Selection for Intrusion Detection System Using a Hybrid Efficient Model
Sivasangari Gopal, M. Sathya · Zenodo (CERN European Organization for Nuclear Research) · 2018
In modern technologies network intrusion detection system plays an important role to defence the network system security. Network traffic and multi-type network system lead to considerable increase of vulnerability and intrusion. Due to the complexity of network attacks, it is always important to achieve high performance security methods, which need to thwart different attacks. Intrusion detecting method identifies whether the network traffic be normal or anomalous from the gathered information that related to security policies. High-dimensional input data analysis is most confronts in IDS. Feature selection frequently encounters this difficulty. This paper proposed the hybrid efficient model used to analyse the optimal features in the data, and it improve the detection rate and time complexity effective. This approach deals with high false and low false negative rate issue, first pre-processed data should be correlation based particle swarm optimization with GR-CR (Gain Ratio & Co-Relation) combination of this approach provide learning based some important subset of features and shows progress in the accuracy and time complexity level. Next, the novel approach tested on KDD cup 99, ISCX and ITDUTM dataset. This approach-achieved machine learning methods and it provide better performance in terms of accuracy rate, time taken, precision, and recall of the networks. Proposed approach compared with the following classification methods: Tree, Bagging, Navie Bayes, RBF classifier, Multiclass classifier, Logistic. The simulation results, gave the high detection accuracy (99.7%) in KDD 1999, (98.3 %) in ISCX and (99.3%) in ITDUTM with fewer feature selection.