Exploring the weightage of correlates of Diabetes prediction using Machine learning

Abstract

There is an increase in the prevalence of Type 1 and Type 2 Diabetes/diabetes around the world. This alarming trend highlights the  need for prediction tools that can help identify risk factors associated with these chronic diseases, so that interventions can be implemented at a much earlier stage in their development. This study investigates two distinct datasets drawn from Kaggle, focusing on clinical and lifestyle factors, respectively. We constructed machine learning models, such as Logistic Regression L1 penalty, Logistic Regression L2 penalty, Random forest and a dummy classifier model to contrast basic accuracy, to first make predictions of the likelihood of diabetes occurrence given various factors. We demonstrate that the models have high predictive accuracy, with the logistic regression L2 penalty model  achieving 95.297% accuracy, the logistic regression L1 penalty model  achieving 96% accuracy, and the random forest model achieving 97% accuracy. However, the key contribution of this study is to provide interpretation of these models to determine the most important drivers of the models’ predictions. We find that  even though well-known factors such as HBA1C level, hypertension, and heart disease have high associations with diabetes, factors such as mental health even though below BMI and HBA1C level do have a moderate predictive power.