Advancing malware classification with hybrid deep learning
Chougdali Khalid, Rabii El Hakouni · 2024
The landscape of cybersecurity is marked by the perpetual challenge of malware classification, demanding adaptable and robust solutions to confront evolving threats. In this research, we present an exhaustive study centered on the innovative realm of hybrid deep learning, specifically exploring the integration of DenseNet and long short-term memory (LSTM) for malware classification. Our hybrid deep learning approach ingeniously blends convolutional neural network (CNN) feature extraction with LSTM sequence modeling. Trained meticulously on a curated dataset comprising six diverse classes of malware, this hybrid model showcases exceptional prowess, boasting a loss of 0.00769, accuracy of 99.73%, precision of 99.53%, recall of 99.65%, and an F1-Score of 99.62%. These metrics accentuate the model’s precision and recall capabilities across an array of malware types. The comparative analysis we undertake unveils valuable insights into the strengths and nuances of hybrid deep learning for malware classification. This research underscores the effectiveness of the DenseNet and LSTM approach, positioning it as a compelling candidate for precise malware classification. As cybersecurity practitioners grapple with the ever-evolving threat landscape, the synthesis of these findings equips them with a potent tool for informed decision-making, highlighting the immense potential of hybrid deep learning in enhancing cybersecurity practices.