Data augmentation using conditional generative adversarial network (Cgan) for android malware binary and multi-class classification

In the field of mobile malware detection and classification, various datasets have been constructed, particularly for Android operating systems, which experience the highest average of malware attacks. Traditional detection and classification methods that rely on malware signatures are becoming increasingly ineffective, especially against zero-day malware, necessitating smart detection methods. Machine learning models have evolved to address this problem as new malware becomes more sophisticated. To improve these models for malware classification and detection, new datasets are required to enhance accuracy. However, constructing these datasets manually is tedious and complicates the prediction of zero-day malware patterns. Additionally, there is a lack of comprehensive Android malware datasets to improve detection accuracy. In this study, a new synthetic multiclass Android malware dataset was generated from one of the most recent datasets, CCCS-CIC-AndMal2020, using two approaches as conditional inputs to a Conditional Generative Adversarial Network (CGAN) model: binary classification and multiclassification. A machine learning classification model was developed to evaluate the results of the two approaches by combining the generated data with the original data to test the quality of the generated data. The multiclassification approach achieved low accuracy, while the binary classification approach achieved 85% accuracy using LightGBM and 78% using Random Forest, compared to the original data accuracy, which was 85% using both LightGBM and Random Forest.