Diabetes Detection by Data Mining Methods

Globally, Diabetic mellitus (DM) is the most common chronic health problem worldwide. It elevates the blood glucose levels in patients, resulting in severe health problems like fatigue, numbness on feet or hands, blurred vision, etc., particularly, type-2 DM prevalence rate is increasing day by day. However, it is a curable disease, so it is vital to identify the disease early to reduce the complications. Correspondingly, traditional screening methods for DM include a blood test, blood sugar meter, etc. It is a costly and time-consuming method. Moreover, it requires physicians to process the test. To resolve this, numerous existing researches focused on the detection of DM with Artificial Intelligence but lacked accuracy and speed. Therefore, the proposed system used a particular set of procedures to enhance the classification performance. Initially, the data is preprocessed to remove the noise and enhance the quality of the data. Then, the feature extraction is carried out using Kernel Principle Component Analysis to take the significant features in the data. Further, feature selection is processed using Novel Bisection Of Branch And Bound Algorithm to select the important features. Formerly, classification was attained through Novel-based Radial basis function based Kohonen Self-Organizing Map with Grey Wolf Optimizer. The performance of the system is verified through performance metrics. The outcome of the analysis signifies the proposed model accomplished an accuracy of 97.07%. The respective system is planned to contribute to the research related to diabetes and to support qualified doctors in diabetes treatment.