MACHINE LEARNING TECHNIQUES FOR DIABETES CLASSIFICATION

Abstract

Driven by the explosion in the generation of Biomedical Data and their complexities, Machine learning approaches have been found to be extremely compelling for detection, diagnosis and necessary medical decision making of diseases. The objective of this paper is to investigate the efficiency of various Machine Learning based algorithms in the analysis of very common and fatal disease like Diabetes. These algorithms are not only classifying the diabetic patient into different categories but also, they are advising the diabetic patient suffering from other associated diseases (originated due to diabetes) for immediate medical attention or not. The two datasets used in the study are Pima Indians Diabetes Dataset and 130 US Hospitals Data for the Year 1999-2008. The various Machine Learning algorithms used in the study include Logistic Regression, K- Nearest Neighbors, XGBoost, Decision Tree, Random Forest, Support Vector Machines and Neural Network based MLP Classifiers. The efficacy of the models is tested on the basis of Classification Accuracy and F1 Score. The results are analyzed and compared. It demonstrates that Logistic Regression model outperforms other models in the study of Pima Indian Diabetes Data whereas Neural Network based MLP Classifier outperforms other models in the study of the Diabetes 130-Us Hospitals Data for Years 1999-2008.