Hybrid Diabetes Mellitus Prediction Modeling and Analysis Using Machine Learning Techniques

Diabetes happens when your body fails to manage the portions of the foods you take to convert them to energy. Diabetes, therefore, needs to be detected and treated early well in advance so that health complications may be averted. More recently with the methods of machine learning in place, diabetes can be better predicted. This research will propose a novel approach to a patient‘s diabetes risk level based on a combination of the traditional clinical and the newly proposed HL7 insurance profile. The specific algorithms evaluated and compared are random forest, logistic regression, and ensemble of the two approaches, which when combined, would improve upon the current state-of-the-art methods for the prediction of diabetes risk. In this work, model was evaluated on benchmark datasets containing 15 attributes and the performance was evaluated using the ‘n-fold cross validation‘. For training purposes, the data base was developed using 1500 healthy people from the general population selected at random. The hybrid model has been evaluated with LightGBM. The accuracy of the hybrid model with the LightGBM is 76% for the collected dataset, and 90% for the Pima Data set when applying the problem instances attributes.