JoWUA

Abstract

Medical diagnosis is usually associated with uncertainty and vagueness, as well as inconsistent clinical information, which decreases the classification reliability. To surmount this problem, this research suggests that the Intuitionistic Fuzzy Adaptive Decision Tree (IF-ADT) that incorporates the dynamic threshold optimization method with intuitionistic fuzzy sets (IFS) can maximize diagnostic accuracy in case of uncertainty. The given approach adds membership, non-membership, and hesitation levels to the decision tree splitting process, allowing updating the threshold adaptively due to statistical variance in patient characteristics. It tested the model on 10-fold cross-validation on ten benchmark UCI medical datasets. Experimental performances have shown that IF-ADT has better performance than traditional classifier(s) like Bagging, J48, Naive Bayes, and Multilayer Perceptron. In particular, IF-ADT was 99% accurate and 99% sensitive to chronic kidney disease, 98% sensitive and 98% sensitive to Breast Cancer, and had a specificity of over 95% on most datasets. The mean accuracy increase in comparison to the traditional decision trees was about 3-5 %, and the false positive rate decreased by approximately 4 %. These findings verify that the combination of dynamic thresholding and intuitionistic fuzzy reasoning is very useful in increasing diagnostic robustness and interpretation. The suggested IF-ADT model is thus applicable in intelligent clinical decision support systems that need a great degree of reliability in questionable medical circumstances.