Comparison of tree-based machine learning methods and its application to diagnosis

The aim of this study is to investigate the potential of using tree-based machine learning algorithms in the diagnosis of thyroid disease. For this purpose, the chronic autoimmune thyroid dataset collected at Muğla Sıtkı Koçman University Medical Faculty Hospital and the open dataset from the University of California Irvine Machine Learning Repository were used. Synthetic minority oversampling technique (SMOTE) was applied to the imbalanced chronic autoimmune thyroid dataset. Five different classification methods were applied to the data sets. The applied classification methods are C5.0 decision tree algorithm, CART decision tree algorithm, CTREE decision tree algorithm, random forest and xgboost methods. Open source R programming language libraries were used in the application. The confusion matrix was used to evaluate the performance of the classification methods based on accuracy, confidence interval for accuracy, sensitivity, precision, specificity, F score value and kappa statistics. In the light of these results, the techniques that gave the best results are specified. In addition, the models were evaluated in terms of decision rules. Thus, methods that are both high-performing and simple to interpret were examined.