A Hybrid Approach for Diabetes Prediction and Risk Analysis Using Data Mining

Abstract

With the increase in patients suffering from diabetes, early detection and prediction of diabetes are the major area of concern. In this study, we propose a hybrid model using data mining techniques to analyze the available data to predict the occurrence of diabetes. This model is a combination of cluster and class-based approach which uses K-means and weighted K-means for clustering and logistic regression for classification. K-means is a simple and widely used technique, but it is highly sensitive toward initial centroids and outliers which further affect the prediction accuracy of logistic regression. The aim is to determine a way to improve the initial centroid selection for K-means and retain maximum original dataset to enhance the performance of logistic regression. Results show that accuracy of the classification model using K-means and weighted K-means is 96.97% and 97.84%, respectively. Further, using the classification results, this paper analyzes the risk associated with diabetic and non-diabetic patients.