A Hybrid Deep Learning Based Approach for Malware Detection and Classification

Malicious software, commonly known as malware, takes advantage of anomalies in computer security to cause damage or illegal access. Recent advances in deep learning have enabled the conversion of malware codes into images, opening paths for developing new detection approaches. This translation allows deep learning models to automatically detect and identify minute traits that point to dangerous activity by taking advantage of the robust visual patterns present in the malware structure. This study, the performance of three deep learning models: Convolutional Neural Network(CNN), Recurrent Neural Network(RNN), and Restricted Boltzmann Machines (RBM), on a dataset of 9,338 grayscale photos spanning 25 malware classifications. With an accuracy of 97 % and a precision, recall, and F1-score of 92%, the present work results show that the CNN model outperforms the others. All malware types have a consistent prediction confidence of 98 %, as concluded from the class-wise analysis. This work may include much more reliable cyber-security solutions as it demonstrates that deep learning, and CNN-based approaches in particular, provides a powerful and robust method to detect malware by inspection of its picture representations.