Effective Diagnosis of Diabetes Mellitus using Voting Ensemble of Boosting Algorithms - Distinctive Machine Learning Approach
Chokiyan Karthikeyini, Balaraman Sundarambal, R. M. Bommi, Suresh Subramanian
Abstract
The rapid growth in the integration of machine learning technology with data exploration modules facilitate processing huge quantum of data in healthcare and develop clinical decision-making system for the diagnosis of pathological condition. Such integrated units assist the physicians to determine the abnormality effectively. In this paper, we have proposed a distinctive integrated Diabetes Mellitus Machine Learning Model (DMMLM) for predicting Diabetes Mellitus (DM) efficiently. The variables considered as a measure of severity are age, insulin level, glucose, pedigree function, body mass index (BMI), skin thickness, pregnancy and blood pressure. Three different machine learning prediction models using boosting algorithms such as Adaboost, Gradient Boosting and XGboost along with voting ensemble are employed to understand their efficacy. Pima Indian Dataset from UCI repository that consists of 758 observations with 8 features is used for the present study. The results obtained shows that the three proposed DMMLM with the voting ensemble for identifying DM provide improved prediction values compared to other models suggested for a similar purpose. The proposed voting ensemble classifier outperforms with higher accuracy of 88.9%, precision 90%, Recall 93%, F1 score 91% and AUC of 0.96.