Diabetes Risk Prediction Using the Classification and Regression Tree (CART) model and the Agreement between Indian Diabetic Risk Score and the CART Model with Kappa Statistics

Abstract

Background: Diabetes is a chronic disease that occurs because of an imbalance in blood sugar levels. The prevalence of diabetes mellitus is rising rapidly worldwide. A crucial aspect of preventative medical care is estimating the risk of diabetes. The use of artificial intelligence is increasing in the healthcare industry for the management, prediction, and diagnosis of diseases. Classification and regression tree (CART) is a predictive model in artificial intelligence that can be used for early prediction with better accuracy.

Objective: This research was conducted with the objectives of: a) estimating the diabetes risk by the Indian Diabetes Risk Score b) to predict the diabetes risk using the Classification and Regression Tree (CART) method, and c) to evaluate the agreement of predictions using Kappa statistics.

Methodology: The secondary database of 380 participants from our institutional study of IDRS was used for the prediction of the risk of diabetes.  The classification and regression tree (CART) model were used for the prediction of IDRS risk. The performance of the algorithms was evaluated on the basis of prediction accuracy of classification. Kappa statistics were calculated to check the agreement between IDRS assessment and prediction.

Result: The prediction accuracy of CART classification for the training data set was found to be 98.7%, and that of testing was 95.8%. Weighted Kappa statistics for IDRS risk and CART classification were 0.967, showing almost perfect agreement.

Conclusion: Diabetes can be predicted using the CART approach in the same way as with the traditional IDRS method.