Comparison of Transfer Learning Performance in Image-Based Malware File Classification on the Dumpware10 Dataset
Febri Putra Sandhya Prawiranata, Raden Budiarto Hadiprakoso · 2023
Malware is computer software created with the purpose of damaging system functions, disrupting performance, and stealing information from a system. Efforts to address malware involve antivirus software, which has evolved to utilize signature-based and anomaly-based techniques. However, these techniques face challenges in detecting and classifying polymorphic malware, as this type of malware can alter its data structure with each execution. One approach to tackling polymorphic malware is converting binary malware into image files. This approach demonstrates that polymorphic malware is unaffected when transformed into visualized images. This research focuses on classifying image-based malware files using deep learning, employing several transfer learning models such as EfficientNetV2, InceptionV4, and Xception on the Dumpware10 dataset, which consists of 11 categories of malware visualized as images. Additionally, the research applies data augmentation methods to each model. The study aims to determine the best performance of deep learning algorithms using transfer learning models for classifying image-based malware files in the Dumpware10 dataset. It also investigates the comparison of training times between models with and without the application of data augmentation methods.