Comparison of Fuzzy Diagnosis with K-Nearest Neighbor and Naïve Bayes Classifiers in Disease Diagnosis

Asaad Mahdi, Ahmad Razali, Ali AlWakil


The main objective of this paper is to investigate the performance of fuzzy disease diagnosis by comparing its results with two statistical classification methods used in the diagnosis of diseases namely the K-Nearest Neighbor and the Naïve Bayes classifiers. The comparisons were made using
the latest XLMiner® and Medcalc® statistical software’s. The first step was using fuzzy relation such as the occurrence relation and confirmability relation on a sample of 149 patients suffering from chicken pox, dengue and flu taken from different general and private hospitals and clinics in Kuala Lumpur to diagnose the three diseases. Fourteen symptoms were used in the diagnoses such as high fever, headache, nausea, vomiting, rash, joint pain, muscle pain, bleeding, loss of appetite, diarrhea, cough, sore throat, abdominal pain and runny nose. The second step was using the KNearest Neighbor classification method and the Naïve Bayes classification method on the same sample to diagnose the three diseases. The final step was the comparison between the three methods using performance tests, McNemar and Kappa tests. The result of the comparison between the three methods showed that fuzzy diagnosis outperforms the other two methods in disease diagnosis.


Fuzzy set theory, K- Nearest Neighbor Classifier, Naïve Bayes classifier. McNemar test, Kappa test, performance tests

Full Text:


(C) 2010-2022 EduSoft