Predictive Analysis of Diabetes Mellitus Using Decision Tree Approach

Abstract

Diabetes Mellitus is one of the Non-Communicable Disease. It is characterized by hyperglycemia. It may be reason for numerous health difficulties. According to the statistics in coming years, in 2030, worldwide it reaches up to 643 million counts of diabetic patients. The various factors such as Bad Diet, High Blood Pressure, Obesity, Age, Lack of Exercise etc. can led to Diabetes Mellitus. The various diseases like nerve damage, heart problem, eye problem, kidney disease, stroke in many people caused by high risk of Diabetes Mellitus. Machine Learning plays vital role in healthcare sectors. Using various Machine Learning techniques, a study can be done on large datasets and discover unseen facts, unseen patterns to realize awareness from the data and predict results consequently. In this work, the PIMA Indian Diabetes dataset is considered for Predictive Analysis. The Decision Tree algorithm, namely ID3 (Iterative Dichotomize 3) is used for a predictive analysis. The major phases include preprocessing, feature extraction, classification, and prediction. The classification accuracies of 79.76% and 76.3% are obtained for balanced and unbalanced datasets respectively. The maximum specificity and sensitivity obtained are 79.75% and 79.78% respectively. The ROC curve gives 79.57% overall classification accuracy. The work assists physicians as a second opinion. It can help various health industries to predict the diabetes disease in a better way.