Exploring Automatic Malware Detection Through Deep Learning Models
Jaafar M. Alghazo, David M. Feinauer, Sherif Elmeligy Abdelhamid · 2023
Malware is continually being developed and reinvented by malicious users. Deep Learning opens doors for real-time detection and classification of malware. In this paper, we explore the use of deep learning models (GoogleNet and ResNet50) for the detection and classification of malware based on the Malimg dataset. In addition, we explore modifying deep learning models and present a modified ResNet50 model to study its effect on classification accuracy. Finally, we explore the use of ensemble methods, applying three independent ResNet50 models to study the effect of ensemble models on classification accuracy. GoogleNet achieved the highest test accuracy among the three with 94.82%, however, when applying the ensemble method, the test accuracy reached 97.86%.