DEEP LEARNING FOR CYBERSECURITY : Using a Deep Learning algorithm to predict intrusion

Because of the increased emphasis on cyber security in today's world, intrusion detection systems (IDS) have evolved into a critical component of cybersecurity. IDS promotes the incorporation of Deep Neural Networks (DNNs) for a variety of reasons, including the Covid 19 pandemic and the increased complexity of advanced cyber-attacks. The purpose of this thesis is to explore whether Deep Learning is a usable tool for IDS. In this thesis, DNNs were used to predict Intrusion Detection System (IDS) attacks and the KDDCup 99 dataset was used to train and test the network. A variety of classical Machine Learning algorithms, such as Decision Tree, K-Nearest Neighbor, and Naïve Bayes, to name a few, are used for training. Furthermore, DNNs with one to five layers are used to compare the performances of each model. The findings of the study are presented in a table and show that a DNN with three layers outperformed all of the different DNNs, as well as the Logistic Regression and Naïve Bayes but the model, failed to outperform the Decision Tree algorithm and the K-Nearest Neighbor. Depending on the company’s needs Deep Neural Networks are great models for Intrusion detection systems if the company has the resource, and dataset, if the company only needs the classification of the data and is not interested in gaining knowledge about the data then Deep Neural Networks are favored. Otherwise, if the company needs information a Machine Learning algorithm and especially Decision Trees are still the most usable models available as it not only classify the record but also provide useful information about the data and therefore more information about the possible attacks