Design and implementation of a deep neural network approach for intrusion detection systems

Abstract

Due to the increased reliance of many a human system on cyber infrastructures, there is the continuous need for intelligent protection schemes to be developed as the spate of cyber-attacks is on the rise. Such intelligent systems that could easily detect both existing and zero-day attacks are highly sought in the complex and fast evolving cyberspace. In this regard, machine learning as well as deep learning models have been very effective. However, there is the need for better performing models than the existing ones. This work is aimed at the design and implementation of a Deep Neural Network model for detecting intrusions in computer networks. Techniques such as SMOTE and Random Sampling were applied to handle data imbalance in the CICIDS 2017 dataset. The entire experiment was carried out on a single Jupyter notebook in the Google Colaboratory environment where relevant software libraries such as seaborn, pandas, matplotlib, keras, and Tensor Flow were imported and thereafter implemented as required. Performance metrics of accuracy and loss at both training and validation of the model were considered. Results show that the deep learning model was excellent at predicting attacks with the CICIDS 2017 dataset, achieving accuracy score of 99.68 % and loss of 0.0102.