SYSTÈME AUTOADAPTATIF DE DÉTECTION D'INTRUSIONS INFORMATIQUES VIA APPRENTISSAGE AUTOMATIQUE.

Abstract

These days, computer systems and networks are exposed to computer attacks, for this, intrusion detection systems IDS play the important role in cyber security. The main objective of this study is to realize an IDS in order to secure a Computer Network (SDIR). We are located in a signature-based approach. To do this, we used machine learning by adopting the Naïve Bayesian (NB) method. The results obtained with a classification in two Classes (NB2C), normal and abnormal are better than those of the classification in five Classes (NB5C), normal and 4 attack categories. However, we opted for the first algorithm and we combined it with a method from deep learning which is the recurrent neural network with Short Term Memory (STM). In order to train and test our SDIR, we used the NSL-KDD database. We performed a comparison between the applied algorithms NB5C, NB2CMCLT AND the MCLT, calculating precision, F-Measure and recall. Finally, the results obtained via MCLT alone are better than those of the other algorithms, proving once again the effectiveness of deep learning.