Page 119 - Demo

7A multiclass classification model for tooth removal procedurest117As a classification algorithm, Gaussian Naïve Bayes (GNB) was used. It is a probabilistic machine-learning algorithm, which can be used for a variety of classification tasks. Our dataset has limited size and high variance, with an approximately Gaussian (or normal) distribution. Naïve Bayes classifiers are well known for their performance on problems with a small amount of training data [18], whilst logistic regression models – used for feature selection in this paper - are more prone to overfitting for such problems. Accuracy, precision, recall and F1-score were calculated for each tooth class to evaluate the model performance. To reduce the risk of selection bias and to more accurately estimate the model’s predictive performance, a stratified 4-fold cross validation was performed. In this cross validation, 4 random subsamples of data are used to calculate the performance metrics with the same class proportions (stratified), due to the small sample size. Data availabilityData required to reproduce these findings, are available to download from https://www.doi.org (digital object identifier: 10.4121/19665990).ResultsClinical characteristicsA total of 127 experiments were performed on seven fresh-frozen Caucasian specimens. In 110 (86.6%) experiments, data was successfully recorded. A heterogeneous group of teeth in terms of restorative and periodontal states was included (Appendix Table 2). Feature design In total, 75 features were designed, of which 33 remained after regularization. An overview of these selected features is given in Table 1. The relationship between 2 strong prediction features, the sum of delivered torques and average torques on all three axes is shown in Figure 2. It is an example of how these features can be used to distinguish different classes of teeth. Whilst the sum of torques in all directions can be high for both upper- and lower jaw bicuspids and molars, it seems that average torques in the lower jaw are higher in the dorsal area compared to the upper jaw. Also, in both upper- and lower jaw incisors, average torques did not reach above 6Nm.Tom van Riet.indd 117 26-10-2023 11:59