Unmasking Malware with MDCNN: A New Era of Image-Based Detection

Cybercrime has been a worldwide issue since the internet's inception. The various online crimes that take place have an impact on the general public. The various forms of cybercrime committed online and around the world are always evolving due to the development of the internet. In the current digital age, when malware diversity and volume are increasing rapidly, new methods must be used to identify malware more quickly and accurately. Malware, which includes worms, trojans, spyware, and adware, can have serious repercussions, including financial losses, data breaches, and the interruption of essential services. Manual heuristic inspection in malware analysis is neither efficient nor effective in keeping up with the rapid spread of malware or in analyzing new infections. Deep learning enhances automatic malware variant detection and classification as it provides better categorization by building neural networks with more potentially different layers. They have been applied to automate investigations using static and dynamic analysis, where malware with comparable behaviors is grouped and unidentified malware is categorized into families according to how close it is to the other malware. To analyze malware in a file, certain features from the PE header format may mislead the model into classifying malware as benign. However, when the same file is represented as an image, the distinct patterns in different sections make it easier to categorize the file as either malware or benign. Furthermore, as new malware variants emerge daily, the complexity of these patterns increases, making it difficult for traditional models to analyze them effectively. Our research proposes a deep learning model, MDCNN, to address the challenges faced by existing models in accurately categorizing executable files represented as images into malware or benign. This research study extracts various feature sets from malware image files, including patterns, sizes, and textures, we utilized a novel approach, the Malware Detection Convolutional Neural Network (MDCNN) analyzes the malware in image files and the proposed model shows 96.83% accuracy rate in classifying the malware according to its family. The efficiency of the model is compared with other deep learning models and has shown the proposed model outperforms the other models.