![]() | Only 14 pages are availabe for public view |
Abstract Intrusion Detection System (IDS) is a serious part in any network for security infrastructure. There are different ways to compromise between stability and security of a network. Many computerized methods have been introduced during the last decade to develop intrusion detection systems. A complete security mechanism requires IDS for purpose of monitoring security. A fast reaction to attacks is an essential requirement in network security so, automatic detection system is needed. As earlier as an attack is detected, the more time network administrators have to bring up-to-date their signatures and reconfigure their detection and remediation systems. In this work, combined modules are built to attain better results for the well-known ID metrics as well as introduce a trade-off metrics to characterize each combination. The concept of this proposed ID technique depends on four steps methodology. The first step is to perform data partitioning. The second step is clustering; either using the Possibilistic Fuzzy C-Means clustering (PFCM) technique, or the K-Means Clustering (KMC) technique. The third step is to perform classifying using ANNs, two types of neural networks are used; either Feed Forward Neural Network with back propagation (FFNN) or Radial Basis Neural Network (RBNN) to decide which is better in terms of precision detection. Finally, assessment module is employed to aggregate and assess these results. The experiments are carried out using MATLAB 2015b, and the results using the KDD CUP 1999 dataset show that the proposed approach, PFCM outperforms KMC-ANN, FCANN, BPNN and other well-known methods in terms of detection precision. |