Visual data mining techniques for classification of diabetic patients

Abstract

Clustering is a data mining technique for finding important patterns in unorganized and huge data collections. The likelihood approach of clustering technique is quite often used by many researchers for classifications due to its' being simple and easy to implement. It uses Expectation-Maximization (EM) algorithm for sampling. The study of classification of diabetic patients was main focus of this research work. Diabetic patients were classified by data mining techniques for medical data obtained from Pima Indian Diabetes (PID) data set. This research was based on three techniques of EM Algorithm, h-means+ clustering and Genetic Algorithm (GA). These techniques were employed to form clusters with similar symptoms. Result analyses proved that h-means+ and double crossover genetics process based techniques were better on performance comparison scale. The simulation tests were performed on WEKA software tool for three models used to test classification. The hypothesis of similar patterns of diabetes case among PID and local hospital data was tested and found positive with correlation coefficient of 0.96 for two types of the data sets. About 35% of a total of 768 test samples were found with diabetes presence.