TABLE 1. The Performance of Using Different Strategies Including the Network Size and Initialization for Training CNN Classifiers to Differentiate Between the Two Common Tick Species; Blacklegged vs Dog Ticks. The Best Performances Per Each Column are in Bold and the Second Best Scores are Underlined. ROC-AUC is the Area Under the ROC Curve and PR-AUC is the Area Under the Precision Recall Curve. Regardless of Initialization, CNN Models With Larger Number of Trainable Parameters Perform Better on Tick Data Set. The CNN Classifier Performs Very Poorly if the Initial Layers are Fixed During Training. * Only the Last 5 Layers of the Inception-Resnet Were Fine Tuned While the Rest of the CNN in the Table Were Trained From Scratch Without Any Frozen Layers.
| Model | Initialization | # Trainable Parameters | Accuracy | ROC-AUC | PR-AUC |
|---|---|---|---|---|---|
| lighter CNN | Random |
 |
91.68 ± 0.25 | 97.55 ± 0.34 | 95.43 ± 0.46 |
| Inception-Resnet | Random |
 |
92.04 ± 0.48 | 98.52 ± 0.28 | 96.80 ± 0.99 |
| Inception-Resnet* | ImageNet |
 |
42.10 ± 0.37 | 57.85 ± 0.08 | 47.96 ± 0.20 |
| Inception-Resnet | ImageNet |
 |
91.75 ± 0.06 | 98.51 ± 0.38 | 96.77 ± 0.89 |