Advancing Malware Defense: A Hybrid Approach with Explainable Deep Neural Networks
Deepika Sharma, Manoj Himmatrao Devare · 2024
Malware detection and classification remain critical challenges in cybersecurity, necessitating advanced techniques for combating increasingly sophisticated threats. This paper presents a novel hybrid approach that leverages the power of explainable deep neural networks to enhance malware detection and classification accuracy while providing interpretable results. The proposed method combines static and dynamic analysis techniques to extract a comprehensive set of features from both executable files and runtime behaviors. These features are then processed using a custom-designed deep neural network architecture that incorporates attention mechanisms and interpretability layers. The model not only achieves high accuracy in detecting and classifying various malware families, but also offers explanations for its decisions, addressing the "black box" problem often associated with deep learning models. The experimental results demonstrated the effectiveness of this hybrid approach, showing significant improvements over traditional machine learning methods and baseline deep learning models. The explainable component of the system provides valuable insights into the specific characteristics and behaviors that contribute to malware identification, potentially aiding the development of more robust defense mechanisms. This research contributes to the field by offering a powerful, interpretable tool for malware analysis that can adapt to evolving threats while providing transparency in its decision-making process, thus bridging the gap between advanced AI techniques and practical cybersecurity applications.