We present multiple hybrid methods of analyzing different software categories for machine learning-based malware classification. In the first approach? we introduce a novel image transformation method using statistical? syntactic artifacts? and space-filling curves to convert binary software into color 3D images with semantic information. The second approach is to use static analysis to extract the Interprocedural Control Flow Graph and an additional set of features from a Java bytecode program to produce a grayscale image. We also extract an additional set of features from Java malware programs to improve the accuracy of the malware classification. We evaluate our approaches by leveraging machine learning algorithms? including shallow (XGBoost) and deep (Convolutional Neural Network) learning classifiers for the classification of various datasets of malware. Our experimental results demonstrate that the proposed methods outperformed related works and can detect both known and previously-unseen real-world malicious programs.

Back to OISF 2022 video list