ENHANCEMENT OF ML BASED NETWORK ANOMALY DETECTION SYSTEMS WITH GENERATIVE MODELS

Abstract

A huge rise in network traffic data have brought challenges towards the security of data over network, servers and computers. This has been a challenge to the traditional intrusion detection system also, as the rapid change in technology there changes the intrusion style also. In recent years, new varieties of anomaly attacks have emerged rigorously which can’t be detected by out dated intrusion detection systems. In order to tackle those intrusions we propose a machine learning based approach which implements an autoencoder and a dense neural network. Both autoencoder and dense neural network are types of artificial neural network, but they differ in the processing. Therefore the feature extraction phase of our IDS model is designed on the basis of autoencoder technology and model creation is done on the basis of dense neural network. We also implement a convolutional neural network, another subclass of artificial neural network as intrusion detection system. The dataset used is Intrusion Detection Evaluation Dataset (CICIDS2017). This dataset is of new generation in terms of attacks it contains. The attacks present in the dataset are new types of attacks which are generally used by attackers in real network for the purpose of stealing data. The results obtained by two models are compared i.e. the accuracy in the detection rate is compared and thus after comparing the results the model implemented on dense neural network and autoencoder shows the better accuracy and lesser false alarm rate.