Extreme Morphological Learning Machines for the Detection of Windows Malware

Abstract

Cyber-attacks are causing billionaires losses in ascending order. Then, our goal is to detect malicious behavior of the suspect executable files, preventively, even previously the file be executed by the client. In this current paper, we propose an antivirus to detect malware utilizing mELM (Morphological Extreme Learning Machine) inspired by Mathematical Morphological image processing theory. Our antivirus software is designed to classify executable applications into two categories: benign and malicious. Our mELMs results were compared with classic ELMs and state-of-the-art antivirus programs. All techniques are assessed through broadly utilized classification measurements. Our mELM, through the Dilation kernel, achieves the best result compared to state-of-the-art. The mELM Dilation kernel achieves an average accuracy of 99.76% in the discrimination between benign and malware executable. We claim that mELM can be adapted to any machine learning dataset. Inspired by Mathematical morphology, our mELMs can model any form that exists in decision-making boundary of the neural network.