Implementation of Machine and Deep Learning Algorithms for Intrusion Detection System

Abstract

AbstractThe intrusion detection system (IDS) is an important aspect of network security. This research article presents an analysis of machine and deep learning algorithms for intrusion detection systems. The study utilizes the CICIDS2017 dataset that consists of 79 features. Multilayer perceptrons (MLPs) and random forests (RFs) algorithms are implemented. Four features extraction techniques (information gain, extra tree, random forest, and correlation) are considered for experimentation. Two models have been presented, the first one using the machine learning random forest (RF) algorithm and the second using deep learning multilayer perceptron (MLP) algorithm. The increased accuracy has been observed when using the random forest algorithm. The RF algorithm gives the best results for the four feature selection techniques, thus proving that RF is better than MLP. The RF algorithm gives 99.90% accuracy, and 0.068% false positive rate (FPR) with 36 features. Furthermore, the dimensionality of the features has been reduced from 79 to 18 features with an accuracy of 99.70% and FRP of 0.19%.KeywordsMachine and deep learningRandom forest and MLP algorithmsIntrusion detection systemFeatures dimensionalityCICIDS2017 dataset