Abnormal Files Classification using Convolutional Neural Network CNN

Abstract

As of late, turning Windows files into pictures and analyzing them with deep learning and machine learning have been regarded as state-of-the-art methods for identifying and classifying malware. This is mostly because deep learning model performance in image classification has recently experienced a boom in success, and image-based malware detection and classification is platform independent. Convolutional neural network (CNN) deep learning techniques are successfully used for image-based Windows malware classification, according to a review of the literature. Nevertheless, just a small percentage of the entire picture representation had the infection. Finding and identifying these impacted little areas is crucial to achieving a high degree of malware classification accuracy. This study locates and identifies the little contaminated spots in the overall picture by integrating a Data augmentation technique with a CNN. On a dataset of malware images, a thorough examination and analysis of the suggested technique were conducted. The effectiveness of the suggested Data augmentation-based CNN method was evaluated against many non-attention-CNN-based methods using different data splits from the benchmark malware picture training and testing dataset. The proposed CNN approach ensured computational efficiency and outperformed non-attention-based CNN methods in all the data-splits. Most notably, the majority of the techniques demonstrate consistent performance across all training and testing data splits, illuminating CNN's multi-headed attention. model’s generalizability to perform on the diverse datasets.