A semi supervised classification based approach using autoencoders for intrusion detection system

With the major developments in Internet-of-Things (IoT), cloud-based solutions, and wide selection of data traffic, there is a rising demand for powerful anomaly detection methods for intrusion detection systems (IDS) that can overcome sophisticated network attacks. Rapid advancements in deep learning (DP) and machine learning (ML) have gained the attention of researchers as it has the opportunity to bring a powerful solution to complex attacks. In comparison to other conventional techniques, such as firewall, an Intrusion Detection System (IDS) is the first layer of protection against network-based cyberattacks. Currently available intrusion detection approaches are often based on conventional machine learning models such as Random Forest and Support Vector Machine, and while they depend heavily on manual feature extraction, they have relatively a lower level of accuracy. To overcome the shortcomings of intrusion detection systems in terms of detection and feature extraction, a unique approach to incorporating both supervised learning as well as unsupervised learning (Semi supervised) into Intrusion Detection Systems (IDS) is proposed. The proposed method is utilizing Deep Autoencoder (DAE) being a dimensionality reduction technique. The trained DAE encoder cannot just automatically extract features, also the latent layer can be used as an input to supervised learning algorithms such as Naive Bayes and K-Nearest Neighbor to enhance the detection accuracy of IDS. The overall performance of the proposed model is evaluated using the UNSW-NB15 dataset. The results show that the model has better performance than a number of models where Deep Autoencoder (DAE) is not used.