Comparative Evaluation of CNN, Vision Transformer, and Hybrid Architectures with Grad-CAM Explainability for Vision-Based Malware Classification

Final-year BINUS student capstone, supervised by Gabriel Asael Tarigan. This comparative study evaluates convolutional (ResNet-18), vision-transformer (ViT-Base/16), and CNN-transformer hybrid architectures for vision-based malware classification on the Malimg dataset, with Grad-CAM explanations for interpretability and all models exceeding 97% test accuracy. Note: currently a working draft with placeholder author placeholders.