Parallel Deep Learning with a hybrid BP-PSO framework for feature extraction and malware classification

Malicious software (Malware) is a key threat to security of digital networks and systems. While traditional machine learning methods have been widely used for malware detection, deep learning (DL) has recently emerged as a promising methodology to detect and classify different malware variants. As the DL training algorithm is oriented on gradient descent optimization, i.e. the Backpropagation (BP) algorithm, several shortcomings are encountered, e.g., local suboptimal solutions and high computational cost. We develop a new DL-based framework for malware detection. In this regard, we introduce a hybrid DL optimization method by exploiting the integration of BP and Particle Swarm Optimization (PSO) algorithms to provide optimal solutions for malware detection. Many hybrid DL optimization methods in the literature are not implemented under a parallel computing setup. In this paper, we develop an efficient distributed parallel computing framework for implementing the proposed DL-based method to improve efficiency and scalability. The experimental results on several benchmark data sets indicate efficacy of the proposed solution in malware detection, which significantly outperforms other machine learning methods in terms of effectiveness, efficiency and scalability. • Deep learning with BP-PSO yields very high classification performances in malware detection. • PSO-BP combines the merits of both local–global optimization capabilities. • Parallel computing improves efficiency and scalability of the proposed method in malware detection.