Evaluation of supervised learning algorithms in binary and multi-class network anomalies detection

Information System security is becoming a critical issue today, given the large-scale use of the Internet, the diversity of storage and different means of exchanging information. Solutions developed based on signatures are necessary but ineffective nowadays. The introduction of artificial intelligence has brought new life to the field of network intrusion detection. In this context, through this work, we aim to perform a binary and multi-class classification model using supervised learning algorithms for the prediction of new threats. The proposed approach has been tested on the NSL-KDD dataset. We achieved an accuracy of 80.4% for binary classification and 77.5% for multi-class prediction. These very encouraging prediction rates were obtained with the Support Vector Vachine (SVM) and the Multi-Layer Perceptron (MLP).