Diabetes diagnosis system using modified Naive Bayes classifier

Abstract

In today’s world, Diabetes is one of these diseases and is now a big growing health problem. The techniques of data mining have been widely applied to extract knowledge from medical databases. In this work, a Medical Diagnosis system of Diabetes is proposed for the ‎diagnosis of diabetes in a manner ‎that is rapid and cost-effective. three stages are ‎involved in the proposed diabetes diagnosis system (DDS) including: dataset constructing, preprocessing and classification algorithm using traditional Naïve Bayesian ‎‎(TNB) and modified Naïve Bayesian (MNB)). MNB Classifier is a modified NB that is used to ‎enhance the accuracy of ‎diagnosis, by adding a proposed modest model to help separate ‎the overlapping diagnosis classes. The outcome‎ ‎showed that the accuracy of MNB classifier is generally higher than that of ‎TNB ‎classifier for all sets of features. An accuracy of about (63%) was achieved for the TNB ‎model, whereas ‎that of the MNB model is (100%). The experimental results showed that ‎the MNB is better than the traditional ‎NB in both two cases of constructed medical ‎datasets; the first case of filling the missing values by experiences and ‎the second case of filling ‎missing values by K-nearest neighbor (KNN) algorithm.