Malware Classification Using a Taguchi-based Deep Learning Network

Abstract

Malware is designed to damage computer systems, and malicious targets have proliferated recently. This rising use of malware requires an efficient malware detection method. Because new malware is constantly being created and old malware is constantly updated, manually updating a signature database with newly generated malware samples is increasingly challenging. To reduce the cost of feature engineering and the requirement for domain expert knowledge, researchers have used image-sensing methods to solve the malware family classification problem. In this study, a Taguchi-based deep learning network (TDLN) with optimization of the parameter combination is proposed for malware family classification. A total of 36 experiments were conducted and nine influential factors with various levels were selected for determining the optimal parameters of the proposed TDLN. The experimental results indicate that the accuracy, precision, and recall of malware family classification when using the proposed TDLN are 98.71, 96.90, and 96.78%, respectively. Moreover, the accuracy, precision, and recall of the proposed TDLN are 2.03, 5.59, and 6.09% higher, respectively, than those of the original deep learning network for the Malimg data set.