Application of Different Decision Tree Classifier for Diabetes Prediction: A Machine Learning Approach

Diabetes has created a global impact which has grown dramatically in recent years, making it a global threat. In this study, an effort has been made to predict this silent killer disease using machine learning approach. Benchmark datasets were collected from various countries (Iraq, USA etc.) and five decision tree algorithms—ID3 (Iterative Dichotomiser 3), C4.5, CART, CHAID (Chi-square Automatic Interaction Detector), MARS (Multivariate Adaptive Regression Splines) were applied on the datasets. Decision trees are popular ML technique for easy implementation and interpretability. This study analyzed each distinct approach followed by each decision tree classifier with their merits, demerits, and occasion of application. Several performance parameters (accuracy, precision, recall, etc.) were evaluated and their efficacy in different circumstances is analyzed in detail. The results of the study revealed that diabetes prediction models showed creditable performance rates using decision tree classifier. Even though, CART, C4.5, and ID3 are popular techniques, MARS and CHAID are less investigated. On the other hand, as accuracy is widespread, the significance of recall and precision are many times overlooked. This study addresses these limitations with valuable facts and findings.