Training and Classification Techniques in Intrusion Detection Systems Based on Network Anomalies Comparative Study

Computer network security is vital due to the large volume of data handled. One of the security tools available to large companies is Intrusion Detection Systems (IDS). However, the increase in information and communication technologies has triggered a growth in intrusive accesses and attacks directed at computer systems. This situation has increased over the years, highlighting the vulnerability of such systems. The primary motivation of this research has been implementing the wrapper method applied to IDSs in different training and classification techniques to identify the best intrusion detection model to improve attack detection rates in computer network systems, using a feature selection procedure and other methods of unsupervised training algorithms. In this research, different metrics that measure the quality of the proposed intrusion detection model are evaluated through simulation processes using the DARPA NSL-KDD dataset and by applying the INFO.GAIN feature selection technique to identify the most relevant features in the classification process. Furthermore, an unsupervised learning algorithm (GHSOM, RANDOM FOREST, BAYESIAN NETWORKS, NAIVE BAYES, C4.5, LOGISTIC, PART, AND NBTREE) was trained to classify the bi-class traffic automatically.