SHALLOW CONVOLUTIONAL NEURAL NETWORK WITH EFFICIENT CHANNEL ATTENTION FOR IMAGE-BASED MALWARE DETECTION

Abstract

Malware software, which is designed to malware computer systems, steal personal data, and gain illegal access, is one of the primary cyberthreats. The inability of traditional methods to detect such software has led to the development of more robust and innovative strategies. Imagebased malware detection techniques have become much more common in recent years. These techniques use Convolutional Neural Networks (CNNs) to identify image malware. The aim of the study is to classify malware with a hybrid model combining Shallow CNN and Efficient Channel Attention (ECA) mechanism. The study used a public dataset. Grayscale images in this dataset were converted to RGB color space using a Pseudocoloring technique. The study was evaluated using a 5-fold cross-validation method. The Shallow CNN-ECA model had an accuracy of 0.983. Additionally, with an accuracy of 0.979, the Shallow CNN model ranked second among the suggested techniques. According to experimental results, the proposed model outperformed well-known lightweight CNN methods.