Performance comparison between the proposed RBF SVM, linear SVM, threshold method, threshold voting, and logistic regression. Ten different training/testing splits are used. The mean and standard deviation (SD) of the metrics are shown. TPR denotes true positive rate; FPR denotes false positive rate; PPV denotes positive predictive value
| Algorithm | Threshold Method #1 | Linear SVM #3 | Threshold Voting #3 | Logistic Regression #3 | Proposed RBF SVM #3 | |
|---|---|---|---|---|---|---|
| Feature Set | ||||||
| Accuracy | Mean | 87.9% | 95.1% | 94.5% | 95.5% | 98.4% |
| SD | 2.8% | 1.2% | 1.6% | 1.0% | 0.5% | |
| TPR (recall) | Mean | 89.3% | 91.0% | 77.1% | 94.8% | 93.8% |
| SD | 5.0% | 4.1% | 6.5% | 3.9% | 3.6% | |
| FPR | Mean | 12.3% | 4% | 1.8% | 4.4% | 0.7% |
| SD | 2.8% | 1% | 1.0% | 1.3% | 0.8% | |
| PPV (precision) | Mean | 60.6% | 82.9% | 90.3% | 82.1% | 97.1% |
| SD | 7.8% | 3.7% | 4.6% | 3.1% | 3.6% | |
| F1 score | Mean | 71.9% | 86.7% | 82.9% | 87.9% | 95.3% |
| SD | 6.2% | 2.7% | 3.7% | 2.1% | 1.6% | |