Figure 4.

Algorithm (A) training times and (B) testing times on sample dataset. The dataset is comprised of 2,880 primitives. We computed times to train and test each algorithm on 20–100% of the dataset in increments of 10%. To avoid overfitting and compute an unbiased estimate of training and testing times, ML algorithms were trained and tested de novo with each incremental increase. For training with the complete sample dataset, SVM required the most time (336 s) while the other algorithms finished training rapidly (<30 s). For testing, KNN required the most time (1.5 ms), while the other algorithms finished testing rapidly (<0.03 ms). Please note break in the y-axis to highlight the difference in the algorithm testing times.