Research Article
Android Malware Characterization Using Metadata and Machine Learning Techniques
Table 4
Full benchmark test with top-15 predictors.
| Malware | NumDetectors | F1-score | Precision | Recall |
| Logistic Regression (train/test) |
| 2% | 1 | 0.82/0.1 | 0.76/0.06 | 0.89/0.24 | 25% | 1 | 0.89/0.57 | 0.87/0.46 | 0.91/0.73 | 50% | 1 | 0.93/0.79 | 0.94/0.82 | 0.93/0.77 |
| 2% | 2 | 0.8/0.18 | 0.74/0.12 | 0.88/0.33 | 25% | 2 | 0.9/0.68 | 0.89/0.59 | 0.9/0.79 | 50% | 2 | 0.94/0.82 | 0.95/0.78 | 0.93/0.86 |
| 2% | 4 | 0.81/0.27 | 0.75/0.19 | 0.89/0.47 | 25% | 4 | 0.91/0.73 | 0.9/0.65 | 0.91/0.83 | 50% | 4 | 0.95/0.84 | 0.97/0.79 | 0.94/0.89 |
| Support Vector Machine (train/test) |
| 2% | 1 | 0.85/0.08 | 0.76/0.05 | 0.96/0.23 | 25% | 1 | 0.93/0.68 | 0.92/0.69 | 0.93/0.67 | 50% | 1 | 0.96/0.82 | 0.96/0.87 | 0.95/0.77 |
| 2% | 2 | 0.82/0.16 | 0.72/0.1 | 0.95/0.35 | 25% | 2 | 0.93/0.73 | 0.93/0.7 | 0.93/0.76 | 50% | 2 | 0.96/0.84 | 0.97/0.9 | 0.94/0.8 |
| 2% | 4 | 0.81/0.26 | 0.7/0.17 | 0.97/0.54 | 25% | 4 | 0.94/0.77 | 0.94/0.72 | 0.93/0.83 | 50% | 4 | 0.96/0.87 | 0.98/0.89 | 0.95/0.84 |
| Random Forest (train/test) |
| 2% | 1 | 0.99/0.12 | 0.99/0.07 | 0.99/0.33 | 25% | 1 | 0.99/0.73 | 0.99/0.7 | 0.99/0.77 | 50% | 1 | 0.99/0.84 | 0.99/0.88 | 0.99/0.8 |
| 2% | 2 | 0.99/0.22 | 0.99/0.15 | 0.99/0.46 | 25% | 2 | 0.99/0.77 | 0.99/0.73 | 0.99/0.83 | 50% | 2 | 0.99/0.87 | 0.99/0.89 | 0.99/0.86 |
| 2% | 4 | 0.99/0.32 | 0.99/0.22 | 0.99/0.59 | 25% | 4 | 0.99/0.81 | 0.99/0.76 | 0.99/0.87 | 50% | 4 | 0.99/0.89 | 0.99/0.88 | 0.99/0.9 |
|
|