Enhancing Malware Detection Through Convolutional Neural Networks and Explainable AI

Malware is one of the leading causes of asset and data loss, posing significant risks to individuals and organizations worldwide. Which has recently emerged as a major challenge in the software ecosystem. Early detection of malware is essential to minimize losses. Malware detection is an important aspect of cybersecurity that targets malicious software that threatens computer systems, networks, and data. For this reason, researchers and practitioners have proposed various machine-learning (ML) and deep-learning (DL) models for malware detection. However, with the rapid evolution of malware forms and techniques, traditional identification methods have become less effective at detecting new and unknown malware. In this study, we proposed a convolutional neural network (CNN) model for malware detection from the Malimg dataset that efficiently classifies using less computational resources and without sacrificing accuracy. We employed image preprocessing techniques before training the dataset and we used various performance metrics for this. Our proposed model achieved a test accuracy of 97.49%. We compared our model with six other transfer learning models, such as VGG16, VGG19, ResNet50, DenseNet120, InceptionV3, and MobileNetV2 and derived the test accuracy was 94.04%, 91.87%, 88.59%, 86.78%, and 93.52%. In addition, we used explainable AI (XAI) to determine which features play an important role in the model performance.