Deep Learning Approaches for Malware Detection: A Comprehensive Review of Techniques, Challenges, and Future Directions

Malware remains a persistent and evolving cybersecurity threat, necessitating advanced detection techniques to counter increasingly sophisticated attacks. This survey provides a comprehensive review of deep learning-based approaches for malware detection, synthesizing 109 publications published between 2011 and 2024. We analyze models based on Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and hybrid CNN-RNN architectures, evaluating their performance across publicly available datasets, including BIG2015, Malimg, Drebin, Malicia, and IoT-23. Reported detection accuracies range from 93% to 98.7%, with false positive rates as low as 1.1% in hybrid models. This survey contributes a structured taxonomy diagram categorizing models by architecture and analysis type, an ablation study summarizing individual and combined contributions of CNN and RNN components, and a comparative evaluation of key performance metrics. We further discuss current challenges, such as adversarial robustness and computational complexity, and propose future research directions to guide ongoing advancements in deep learning-based malware detection.