Table 3.
Training parameters and evaluation metrics of successful classifiers used for Phase-1.
|
Phase-1 |
Accuracy (100*Pi) | Total Misclassification Cost | Prediction Speed ( obs/sec) | Training Time | Model Type | Feature Selection |
|---|---|---|---|---|---|---|
| Cosine KNN | 92.2 | 107 | 53 | 877.97 sec | K = 10, Distance weight = Equal | 2000 features |
| Linear Discriminant | 92.4 | 105 | 1500 | 75.487 sec | Covariance structure = Full | 2000 features |
| Bagged Trees Ensemble | 95 | 69 | 5200 | 1092.3 sec | Maximum number of splits = 6901, Number of learners = 30 | 2000 features |
| Medium Gaussian SVM | 97.2 | 39 | 250 | 214.64 sec | Kernel scale = 45, Multi class method = One-vs-One | 2000 features |
| SqueezeNet Deep Learning | 95.86 | 59 | – | 950 min 20 sec | Iteration = 1000, Learning rate = 10-4, Image size = 227*227*3 | – |