Intrusion detection based on Neuro-Fuzzy classification

Computer security is far from being guaranteed due to the scalability of computer networks, the constant evolution of risks and the presence of noisy information. Several solutions were proposed to ensure the integrity, confidentiality and availability of resources, including intrusion detection systems (IDS). The main objective of the current work is first to take advantage of data mining techniques such as normalization, feature selection and eliminating redundancies in order to analyze the huge data like the KDDCUP'99. The second objective is the learning ability of neural networks and the third one is the fuzzy logic reasoning that realizes knowledge wave characters. The proposed model is the Neuro-Fuzzy model precisely the NEFCLASS (Neuro Fuzzy Classification) model of a generic fuzzy perceptron which is in the form of a combination of neuron and fuzzy system networks. This model is characterized by its powerfulness thanks to its large database and rapidity due to its parallel architecture. Moreover, it can be easily updated through the re-learning process following the scalability potential inherited in its architecture. As well as it is distinguished by an intuitive model presented by linguistic rules that are easily understood by the security operator. To test out the adaptability of our approach in detecting unknown attacks we use a test database , namely corrected test that contains new attacks that are not present in the training set and compare the results obtained by the model ANFIS. We prove that our approach based on NEFCLASS model is more powerful in classifying intrusions than the one based on ANFIS model. To highlight the motivation for using the Neuro-Fuzzy Classifier, a comparative study was conducted by using the full 10% KDD among the NEFCLASS model and other supervised classifiers which are ANN and C5.0.