Comparative analysis of machine learning techniques in prognosis of type II diabetes

Abstract

Artificial Intelligence (AI) is now a days gaining immense importance and is becoming a key technology in many fields ranging from banking industry, to travel industry, to communication industry, and to robotic industry. The use of AI in medical diagnosis too is becoming increasingly popular and has been widely used in the diagnosis of tumors, cancers, hepatitis, lung diseases, etc. Numerous algorithms have been designed that help in the process of decision making by analyzing the hidden patterns in previously held information. The main objective of this manuscript is to apply multiple algorithms to a problem in the domain of medical diagnosis and analyze their efficiency in predicting the results. The problem selected for the study is the diagnosis of diabetes. Authors have identified ten parameters that play an important role in diabetes and prepared a rich database of training data which served as the backbone of the prediction algorithms. Keeping in view this training data, authors implemented three algorithms [Naive Bayes, artificial neural networks (ANN), and K-nearest neighbors (KNN)] and developed prediction models. To calculate the efficiency, the results of prediction system were compared with the actual medical diagnosis of the subjects. The results indicate that the ANN is the best predictor with the accuracy of about 96 % which was followed by Naive Bayes networks having an accuracy of about 95 % and the KNN came to be the worst predictor having an accuracy of about 91 %.