Page 121 - Demo

7A multiclass classification model for tooth removal procedurest119Figure 2. Plot of all 110 data points showing the relationship between two features, the AUC of the torque magnitude (sum of torques on all three axes combined) and average torques (on all three axes combined). L = lower, U = upper, AUC = area under the curve, Nms = Newton meter second, Nm = Newton meter, n = numberModel PerformanceA summary of the performance of the model is given in Table 2. On average, the accuracy was 86% in the training set and 54% in the test set (unseen data). The data is presented in two confusion matrices, which show the cumulative results of the four subsamples (Figure 3). In the test set (unseen data), in 104 out of 110 experiments (95%) the correct jaw (upper/lower) was classified. Also, 97 experiments (88%) were either correctly classified or as a neighboring class.Table 2: Performance metrics of the classification model for both training and test set. n = numberSubsample 1 Subsample 2 Subsample 3 Subsample 4 AverageTraining set n=82 n=82 n=83 n=83Accuracy 84% 88% 86% 86% 86%Precision 87% 90% 88% 87% 88%Recall 84% 88% 86% 86% 86ñ-score 85% 88% 86% 86% 86%Test set n=28 n=28 n=27 n=27Accuracy 64% 54% 56% 44% 54%Precision 84% 61% 65% 44% 55%Recall 64% 54% 56% 44% 54ñ-score 71% 53% 57% 47% 57%Tom van Riet.indd 119 26-10-2023 11:59