Comparative Analysis of Naïve Bayes, Nearest Neighbor, and Logistic Regression Classifiers for the Prediction of Diabetes

Machine Learning algorithms have been known for their simplicity and easier implementation, and they have been used predominantly for disease prediction, diagnosis, and decision-making systems to improve healthcare services. These mechanisms are easier to implement than Deep Learning techniques, require fewer resources, and can be easily integrated with other healthcare applications; hence, they have gained popularity recently. In this paper, a comparative analysis is performed with the most widely used Machine Learning classifiers: Naïve Bayes, Nearest Neighbor, and Logistic Regression for the prediction of diabetes. This experiment is performed with 2 diabetes datasets (National Institute of Diabetes and Digestive and Kidney Diseases and Diabetes prediction dataset) from Kaggle and 1 diabetes dataset from Mendeley data using the MATLAB Online Tool. This paper explores the challenges such as the variables being relatively independent stating the class label in Naïve Bayes classifiers and 10 models' performance evaluation with these 3 different datasets for choosing the best model. Developing adaptive methods for selecting optimal $k$ and distance metrics based on the dataset characteristics to improve KNN's accuracy and performance. The no. of the population suffering from metabolic disorders is growing day by day and relevant diseases are increased at an early age indicating the urgency of early diagnosis and timely treatment to prevent further damage. Diabetes is one of the leading diseases in India including in many developed countries, hence this disease diagnosis is pertinent to the prognosis of patients across the globe. Timely diagnosis can benefit the patients from developing other disorders such as heart diseases, obesity, and renal diseases.