Smart Cyber Threat Detection and Prediction Using Machine Learning

Abstract

Cybercrime is now being committed at an alarming rate due to the rise of digital communications, cloud computing and networks connecting devices across the globe. Modern cybercrime is much more difficult to find than previously and is evolving rapidly and dynamically. Traditional IDSs use mostly "signature based" techniques, which work well to recognize attacks if they have been identified before. However, new and innovative methods of cybercrime can be difficult to recognize by traditional IDS's. This problem can be solved by using a Machine Learning Based Cyber Attack Prediction and IDS Framework. In this research, we will develop a machine learning based cyber-attack prediction and intrusion detection framework using the NSL-KDD Dataset. It uses a Convolutional Neural Network (CNN) to automate feature extraction from network traffic data without having to manually engineer thousands of features. This CNN Model was developed to efficiently classify network traffic into either Normal or Malicious classes. We used this CNN Model to train and test against four types of attacks in the NSL-KDD Dataset (Denial of Service (DoS), Probing, Remote to Local (R2L), User to Root (U2R)).We also applied various forms of data preprocessing (normalization, one hot encoding, and feature transformation) to enhance the quality of the output of our model and increase the stability of our models during training. Our experiments showed that this CNN-based IDS detected malicious activity with a higher degree of accuracy than other approaches that relied on machine learning rules and/or signatures. It was able to quickly and reliably predict potential threats in real time. This method enhances the ability of intrusion detection systems to recognize known and unknown threats.