Fuzzy Clustering Tree Algorithm for Medical Data Classification

Class imbalance is a common problem in medical data classification where uneven prevalence of certain features or diseases causes the learners to be biased toward the majority class. Another common issue is ambiguous samples where multiple identical feature vectors are labelled with different classes. To mitigate these issues, previous works have employed clustering or data resampling techniques at the cost of possible information loss. In this paper, we propose a classification method called fuzzy clustering tree classifier (FCTC) which is a semi-supervised, tree-based extension of fuzzy C-means (FCM) to address class imbalance and ambiguous samples without requiring data resampling. Tested on 8 well-known medical datasets, the proposed method achieves high overall classification performance with the presence of class imbalance and ambiguous samples. In addition, the proposed method provides good interpretability as it enables generation of logical rules from the learned tree.