A model fusion approach for severity prediction of diabetes with respect to binary and multiclass classification

Diabetes Mellitus has impacted millions of people across the globe and continues with the same. It is caused due to increased blood sugar levels, as the pancreas fails to produce enough insulin and if produced, is not able to perform its function, thereby leading to excessive infiltration of sugars into the bloodstream. Seeing the ever-increasing risk associated with this disease, it is evident that a concrete solution for accurate timely prediction of diabetes for its necessary treatment is the need of the hour. In this paper, we have come up with a model for the timely prediction of diabetes mellitus as well as its severity level, using Machine Learning (ML) approaches. This work incorporates both binary as well as multiclass classification of diabetes, i.e., classifying patients into diabetic and non-diabetic, and further classifying diabetic patients into Type-1 and Type-2 diabetic patients. A hybrid model is developed using the technique of model fusion, by combining ANN, AdaBoost, and RF, in series with a Logistic Regression Classifier, and comparing this model with the conventional models, like Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), AdaBoost, etc. The overall accuracy obtained for the binary classification hybrid model is 97% for the first dataset and 79% for the validation dataset. Similarly, in the case of multiclass classification, the hybrid model gave an accuracy of 99% for the first dataset and 89% for the validation dataset. This clearly showed that the hybrid model performed better as compared to the existing conventional models in all aspects. Moreover, diabetic patients can also be classified as either having Type-1 or Type-2 diabetes. In addition to this, the severity level is predicted for the diabetic patients, as per a severity index table, formulated based on their respective glucose levels and age.