| While there are diverse and well-established machine learning algorithms for solving classification problems, there is a need for an algorithm tailored to medical diagnosis. This paper presents an approach for constructing a classification decision tree using Youden's J statistic instead of standard node-splitting criteria such as Gini impurity, entropy, and variance-based approaches. The tree is trained by fitting it to a training dataset and deriving Bayesian probabilities, which are then used to predict the diagnosis for a given set of symptoms. The proposed approach is compared with related machine learning algorithms, including decision tree classifiers, boosting algorithms, and naive Bayes classifiers, and evaluated using two open-access machine learning databases. The implementation of the proposed machine learning algorithm in Python is made available as open access. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.