Performance comparison between the random-forest algorithm, logistic regression, linear SVM, decision tree, neural network, RBF SVM, and SVM-RBF SVM. Ten different training/testing splits are used. The mean of the metrics is shown. FPR denotes false positive rate; TPR denotes true positive rate; PPV denotes positive predictive value. Dual and triple denotes dual- and triple-frequency signals.
| Algorithm | Logistic Regression | Linear SVM | Decision Tree | Neural Network | RBF SVM | SVM-RBF SVM | Random Forest | |
|---|---|---|---|---|---|---|---|---|
| Accuracy | Dual | 94.0% | 95.9% | 96.8% | 96.5% | 97.8% | 93.3% | 98.4% |
| Triple | 96.4% | 97.7% | 96.0% | 98.8% | 99.0% | 93.5% | 99.0% | |
| FPR (False Alarm) | Dual | 2.6% | 2.1% | 1.6% | 2.1% | 1.1% | 0.4% | 0.5% |
| Triple | 0.6% | 0.5% | 0.7% | 1.0% | 0.8% | 0.3% | 0.5% | |
| TPR (recall) | Dual | 86.8% | 92.8% | 94.8% | 93.6% | 96.8% | 75.9% | 95.5% |
| Triple | 95.0% | 98.8% | 96.1% | 98.0% | 99.1% | 77.0% | 97.6% | |
| PPV (precision) | Dual | 87.2% | 89.7% | 92.1% | 89.8% | 94.7% | 97.4% | 97.7% |
| Triple | 94.5% | 95.0% | 96.4% | 95.2% | 96.2% | 98.2% | 98.0% | |
| F1 score | Dual | 86.9% | 91.2% | 93.3% | 91.6% | 95.7% | 85.3% | 96.5% |
| Triple | 94.7% | 96.8% | 96.2% | 96.6% | 97.6% | 86.3% | 97.8% |