Mitigating Cyberattack with Machine Learning-Based Feature Space Transforms
Ayoade, Gbadebo Gbadero
MetadataShow full item record
With the increase in attacks on software systems, there is a need for a new approach in software defense. In this work, we explore machine learning-based approaches for the detection and mitigation of attacks. A machine learning-based approach can help to learn patterns that can be used to detect future attacks. However, due to limited labeled training data in cyber security domain, machine learning-based approaches perform poorly in attack detection. This work overcomes these challenges by leveraging two main methods. First, we leverage deception-based honey-patching to label the attack data that can then used to train machine learning models to detect future attacks. The second method uses transfer learning method to adapt data from domains with sufficient labels to train our models and test on data from a different domain. By leveraging these methods, we can show an increase in the detection performance of machine learning-based models in cyber attack defense systems.