Predicting Diabetes Using Machine Learning: A Comprehensive Framework with Model Interpretability | Chapter 4| New Horizons of Science, Technology and Culture Vol. 3

 

This chapter explores the construction of a detailed machine learning (ML) framework for predicting diabetes using diverse real-world datasets. The alarming rise in diabetes prevalence globally and particularly in developing nations such as India necessitates innovative approaches for early detection and intervention. Traditional diagnostic techniques, though clinically established, often fall short in scalability and adaptability. This study focuses on bridging this gap by integrating ML methodologies that not only offer superior prediction accuracy but also provide transparency through interpretability tools.

 

Key contributions of this work include the comprehensive data preprocessing steps (missing value treatment, normalisation, encoding, and SMOTE-based class balancing), the comparative evaluation of three widely used classifiers (Logistic Regression, Random Forest, and XGBoost), and the use of SHAP values for enhancing model interpretability. Among the models tested, XGBoost achieved the highest performance with an accuracy of 97.93%, AUC of 0.9974, and excellent sensitivity and specificity values, confirming its suitability for real-world healthcare applications. The chapter concludes with discussions on model performance, interpretability, clinical relevance, limitations, and avenues for future research.

 

Author(s) Details

 

Mounika Panjala
Department of Statistics, Osmania University, Hyderabad-7, Telangana, India.

 

Bhatracharyulu N.Ch.
Department of Statistics, Osmania University, Hyderabad-7, Telangana, India.

 

Please see the book here:- https://doi.org/10.9734/bpi/nhstc/v3/5950