Diabetes prediction and validation model using ML classification algorithms

Abstract

Diabetes is now a global wide concern, which can critically impact and disrupt the normal lifestyle and the everyday activities of any individual. Due to the lack of insulin and high glucose content in the body, anyone can get diagnosed with diabetes. Apart from all the medical factors, there are few additional non-medical factors in an individual’s daily life like hypertension, heredity, daily standard activity, smoking habits, body mass index etc. that might play a part in triggering diabetes. Several medical studies reveal that for women sometimes pregnancy frequencies or any kind of heart issues can also trigger diabetes. The paper aims to predict the most critical factor that contributes in triggering diabetes in any individual by using classification and predictive analysis algorithms. Five well known machine learning classification algorithms are used where a filtering scheme based on 75% threshold accuracy rate is employed followed by verification using AUROC metric aiming low error rate and high prediction accuracy. Additionally, the model used Ensemble learning to make predictions and validates the proposed scheme against PIMA Indian Diabetes dataset.