An Intelligent Diabetes Predicting Model for Diverse Ethnicities

Diabetes is a metabolic disorder comprising high glucose level in blood over a prolonged period in the body as it is not capable of using it properly. Diabetes is a major cause of blindness, kidney failure, heart attacks, stroke, lower limb amputation, retinal damage, and foot ulcers. The condition is a result of the inter-linkage of lifestyle choices, xenogenetic, psychological, socioeconomic, medical disorders, and geographic attributes. Machine learning-based decision support systems for the prediction of chronic diseases have become immensely popular for better prognosis/diagnosis support to health professionals. Current computational methods for diabetes diagnosis have some limitations and are not tested on varied datasets or people from different countries which limits the practical use of prediction methods. This study identifies classifiers which work with optimal accuracy over three ethnicities. Three unique datasets were identified for this study which are an Indigenous population of USA, European population, and South Asian population for accurate prediction, diagnosing, and treatment of disease. Machine learning algorithms were applied on the datasets, and a comparative study was made. For South Asian ethnicity, GPC, RF, DT predicted with accuracy of 91.62% each. For European ethnicity, the same was performed with 97%, 98.2%, and 97.8%, respectively. For Indigenous Tribe of USA when GPC, RF, and DT were applied, the performance was 61%, 78.6%, 71.8%. SVM and LDA performed better with 80.2% for Indigenous Tribe of USA. Random forest performed with high accuracy on South Asian and European population and comparable accuracy for tribe of USA. Our study provides a base for reducing the gap in polygenic risk prediction accuracy.