Malware classification using deep learning models

Abstract

This project explores the effectiveness of fine-tuned Deep Convolutional Neural Network (DCNN) models for the classification of malware. Given the increasing threat of malware and the importance of accurate detection, the study leverages transfer learning to adapt pre-trained DCNN models, originally trained on large-scale image datasets, to the task of malware classification. The research specifically focuses on five prominent DCNN architectures: Xception, ResNet50, DenseNet-169, InceptionResNetV2, and VGG16. These models were evaluated using two distinct malware image datasets: the Blended dataset and the Malevis dataset. The methodology involved several key steps: preprocessing the malware datasets images, fine-tuning the DCNN models to adapt them to the specific characteristics of the malware images, and applying techniques such as early stopping to optimize model training and prevent overfitting. Performance metrics such as accuracy, precision, recall, and F1-score were used to rigorously assess the models' classification capabilities. The primary objective of this study was to determine the most effective DCNN architecture for malware classification, providing a comparative analysis of their performance. The findings aim to contribute to the field of cybersecurity by demonstrating the potential of advanced deep learning techniques in improving malware detection systems. By identifying the strengths and weaknesses of each model, this research offers valuable insights for the development of more robust and accurate malware classification tools.