Android Malware Detection Based on CTGAN and LightGBM

Android OS vulnerabilities necessitate improved malware mitigation. Machine learning (ML) offers solutions but struggles with imbalanced datasets, reducing effectiveness in detecting minority classes. Traditional methods include sampling, feature engineering, data augmentation, and ensemble learning. This study proposes using Conditional Tabular GAN (CTGAN) to generate synthetic data for minority classes. Preprocessing includes scaling, classification with LightGBM, and feature selection (FS) to assess its impact. Using the CICMalDroid2020 dataset, results showed CTGAN without FS achieved the highest accuracy (95.17%), followed by FS without CTGAN (94.87%), no FS or CTGAN (94.74%), and both CTGAN and FS (94.66%). These results highlight CTGAN's superior impact on performance.