Diagnosis of Diabetes Mellitus using Hybrid Techniques for Feature Selection and Classification

Abstract

Diabetes is specified as the most chronic and deadliest disease that results in increasing blood sugar. The medical data mining approaches were utilized for detecting unobserved patterns in the medical fields of sets of data for medical diagnosis and treatment. Data classification for diabetes mellitus is quite significant. Two main measures are used in this analysis to differentiate between those with diabetes and others who are not. To find the most effective attributes for this disorder, the first move is to use Hybrid feature selection to find the most effective attributes. Classification is done in the second stage using "Logistic Regression and K-Nearest Neighbors algorithms". Where utilizing two types of data sets, the first is local, collected from consulting laboratories at Baquba General Hospital, and the second is global, which is the Pima India Diabetes Database (PIDD).The experiment on the Local dataset shows that LR without Hybrid feature gives an accuracy of 96%, while with Hybrid feature give accuracy of 98%, KNN without Hybrid feature give accuracy of 90%, while with Hybrid feature gives accuracy of 98%. The experiment on the Pima dataset shows that LR without Hybrid feature gives an accuracy of 76%, while with Hybrid feature gives accuracy of 90%, KNN without Hybrid feature gives accuracy of 81%, while with Hybrid feature gives accuracy of 85%.