Data Mining Techniques for Early Prediction of Diabetes Mellitus

Abstract

The prevalence of diabetes mellitus (DM) as a disease is considered to be a leading cause of mortality rate in the world and tends to effect eye sight, kidney and heart of a human body. Since its occurrence triggers diversity in health issues; detecting it at the right phase is mandatory. For this reason, multiple research scholars have contributed their work in this field of study by adapting to various data mining techniques and thereby reducing the overall workload of medical practitioners. Detection of diabetes mellitus using such techniques have also resulted in its early diagnosis and thereby enhanced the overall treatment of detection at the right stage. The phenomenon of diabetes mellitus can however be categorized as Type 1 and Type 2 diabetes; wherein Type 2 diabetes is responsible to cause heart diseases. Therefore, the primary aim of the research study is to detect the occurrence of DM by utilizing techniques of data mining. For this reason, the authors have implemented the back propagation technique to classify whether an individual is diabetic positive or not. In addition to the back propagation technique, the authors have also implemented Naïve Bayes, Random Forest and J48 with input of neural networks having 8 parameters. The execution of these algorithms is performed using 6 hidden layers of neuron on the PIMA dataset and is further applied on R studio. Throughout the implementation, it has been observed that the back propagation technique generated highest accuracy in comparison to the working implementation of Naïve Bayes, Random Forest and J48.