INTRUSION DETECTION SYSTEM USING DEEP LEARNING

Abstract

Intrusion detection systems should be powerful and reliable in the age of the Internet of Things to ensure the security and integrity of interconnecting devices. In this thesis, we employ deep learning augmentation techniques using Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM) networks. We tested the performances of the models on three benchmark datasets: NSL-KDD, UNSW-NB15, and CICIDS 2017. Our focus was on the ability of the models to classify data into normal and attack classes. We show in this work that both models are highly efficacious, though with some variation in the performance metrics across different scenarios. In the case of data sets, the BiLSTM model outperformed the LSTM in most metrics, with accuracies of over 98% in all cases and an excellent result in the UNSW-NB15 dataset of over 99%. This comparative analysis not only allows us to know the potential of the LSTM and BiLSTM models in the domain of IoT IDS, but also their operational strengths and weaknesses across diverse attack scenarios, which could guide further research and practical implementations of deep learning based IDS in the enhancement of IoT security.