Malware Images Visualization and Classification with Parameter Tunned Deep Learning Model

Abstract

Malwares can be termed as a malicious program that can gain unauthorized access to the computer. This unauthorized access can damage and harm computing world in many capacities. There are many malware detection approaches present in the world. These approaches include static and dynamic analysis, machine learning, semi -supervised and deep learning-based models. These approaches cannot be visualized, thus cyber security experts face difficulty in interpreting underlying patterns. Conversion of malware byte code into images exits. An improved approach that can not only visualize malware, but also predict malware with high accuracy can be beneficial. For this purpose, we have used existing malware visualization technique. A technique which converts malware samples into images and then applies a contrast-limited adaptive histogram equalization algorithm to enhance the similarity between malware image regions in the same family. After conversion into images, we have applied parametrized tunned Convolutional Model to predict malware images. Comparing with existing our approach not only visualizes malware images but also outperforms previous approach by almost 2%, by achieving 98.27% accuracy.