Multi-Model Deep Learning Ensemble Approach for Detection of Malicious Executables
Mohammad Eid Alzahrani · Islamic University Journal of Applied Sciences · 2023
Due to the growing significance of the Internet in many facets of our lives, the World Wide Web, which end-users access via web browsers, is evolving into the next platform for those who want to engage in illegal activity for either their own or another person's financial or personal benefit. Among the reported types of attacks, attacks through malicious executables files are still one of the prevalent challenges. Different static and dynamic analysis approaches have been proposed to detect such executables. The challenge with these approaches is that they failed to detect novel attack types in malicious executables. With the dawn of Machine learning, the detection of novel attacks in malicious executables was possible to detect with high accuracy. Deep learning, which is a part of machine learning that works similarly to human neurons, provides a way to achieve much greater accuracy compared to machine learning. In this study, we propose a stacking-based ensemble approach combining CNN, LSTM, and GRU models to detect malicious executables. The experiment results demonstrate that an accuracy of 99.02% was achieved, which is very high compared to individual deep-learning models.