A Comparative Analysis of Deep Learning Models for Malware Detection Using Image-Based Features

Abstract— Nowadays, malware has been more advanced, trickier, and complex; as a result, we need reliable and scalable detection tools that work across all sorts of threats. Recently, some advancement has been made in image-based malware representation, where threats can be analysed as visual patterns and deep learning models can be used to analyse those patterns, which can be difficult for traditional analysis techniques. In this paper, a systematic comparative analysis of five different deep learning models, such as CNN, VGG16, ResNet-50, DenseNet-121, and EfficientNet-B0, is done for sorting malware into multiple classes using RGB representation. We tested all these 5 models on a public benchmark dataset with matching pre-processing, splits, training setups, and different evaluation metrics to keep it fair and repeatable. Covering lightweight task-specific networks to deep pre-trained models, this study provides hands-on insights for picking and benchmarking models in malware image analysis in cyber defence. Keywords—Malware detection, Deep Learning, Convolutional Neural Network, Malware Image Classification, Cybersecurity, Confusion Matrix, False Negative Reduction