Optimized stacking ensemble models for the prediction of diabetic progression

The influence of applied machine learning in our day-to-day life has seen significant improvement over the last few years. The use of machine learning in Artificial Intelligence to predict various aspects of human life has helped industries in knowledge discovery, to draw inferences and to ultimately increase the business aspects. In healthcare industry, when different machines which monitor various health parameters are increasingly getting connected, it is important to process the information and draw inferences which could be very helpful and easy for the doctors to prescribe medicines and to give advice on lifestyle changes. In this paper, disease progression of Diabetes Mellitus of 442 patients is analyzed in terms of various health parameters along with six related blood serum measurements. Here, optimized stacking method is used to perform both regression and classification. In regression, the quantitative measurement of disease progression is predicted where as in classification, the disease progression is classified into high progression or low progression category. In both cases, certain base models are chosen and the accuracy score of these base models are compared with the score of optimized stacking based ensemble model.Optimized Stacking has shown promising results in comparison with the individual methods. The method is also tested on standard datasets. The result validation is performed using a large dataset with 22 features and 70,692 records, which is used to predict the diabetic information of patients. It was found that the technique has performed well with all the datasets.This method can be used as a data analysis backbone of healthcare based IoT systems for predicting diabetic progression as well as for any other related applications.