Prediction of Diabetes Mellitus using Ensembled Machine learning Techniques

Abstract

Introduction: Machine learning can be used in the prediction of Diabetes Mellitus thus helping to reduce mortality and morbidity arising out of it. The health system in India is already overburdened and the occurrence of pandemic COVID 19 mandates better predictive outcomes with less interaction in the preliminary stage. Methodology: The study uses 768 records of PIMA Indian Diabetes dataset (500 diabetic and 268 non diabetic records) with 8 attributes of each. The data was pre-processed by replacing the missing values with mean and the performance of the model was increased by removing the imbalance in the dataset by Synthetic Minority Oversampling Technique(SMOTE) algorithm.A new ensembled algorithm SDS (using SGD Classification, decision tree and Gradient Boosting) is presented in the study. Results:The accuracy score of proposed SDS algorithm in the test dataset is 73.38%. and the AUC(Area under curve) value was 70.42%. Conclusion: The proposed ensembled algorithm of Diabetes prediction can be used to support the already overburdened health system.