| Summary: | A reliable medical decision-making is essential to diagnose a disease. This assists medical practitioners to detect a disease at early stage especially diabetes that causes further health complications. The diversity and availability of healthcare datasets supports medical practitioners to use computer applications in the diagnosis process. There are many medical datasets available for research usage but these datasets lacks information that allows decisions to be made accurately, which have a major impact to diagnose a disease. Fuzzy logic has contributed to handle vagueness and uncertainty issues and one of the appropriate models for the development of medical diagnostics. Most computer applications use machine learning and data mining techniques to aid classification and prediction of a disease. Therefore, a fuzzy model based on machine learning and data mining is a vital solution. In this study, ten supervised machine learning algorithms namely the J48, Logistic, NaiveBayes Updateable, RandomTree, BayesNet, AdaBoostM1, Random Forest, Multilayer Perceptron, Bagging and Stacking are applied for a simulated diabetes fuzzy dataset, verified by medical experts. The fuzzy datasets provide adequate information on the type of diabetes diagnosis and level of care related to the type of diabetes diagnosis. All algorithms were compared based on the accuracy, precision, recall, F1-Score, and confusion matrix. Experiment results for diabetes diagnosis dataset indicate 100% accuracy for the eight algorithms except AdaBoostM1 which produced 79.82% accuracy and Stacking 67.89% accuracy. In addition, level of care dataset reveals the highest accuracy of 97.15% for MLP and Bagging algorithms and the lowest accuracy of 91.66% for stacking algorithm. Overall, the proposed fuzzy rule-based diabetes diagnosis and level of care fuzzy model works well with most of the machine learning algorithms tested. Therefore, the proposed fuzzy model is a useful aid in the decision-making process, specifically in the healthcare sector. © Little Lion Scientific.
|
|---|