Advancing Multi-Class Diabetes Risk Prediction using Machine Learning, Deep Learning, and Explainable AI: A Data Analytics Approach to Imbalanced Public Health Survey Data

Arshanapalli, Dikshitha (2025) Advancing Multi-Class Diabetes Risk Prediction using Machine Learning, Deep Learning, and Explainable AI: A Data Analytics Approach to Imbalanced Public Health Survey Data. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (861kB) | Preview
Preview	PDF (Configuration Manual) Download (949kB) | Preview

Abstract

Early identification of diabetes and prediabetes which is essential to prevent long term complications and improve population health outcomes. However, most previous studies used binary prediction and limiting the ability to detect intermediate risk states. This research develops a comprehensive multiclass diabetes prediction framework using CDC BRFSS 2015 dataset includes preprocessing, engineered glycaemic features, composite health scores and interaction-based features to enhance predictive signal. Class imbalance was addressed using SMOTE, class weighting balanced boosting. LightGBM, Tuned LightGBM, CatBoost, Deep Neural Network (DNN), 1D Convolutional Neural Network (1D-CNN), and ensemble methods (stacking and soft voting) were implemented and compared using evaluation using accuracy, macro F1, weighted F1, and per-class recall shows that baseline LightGBM delivered better overall performance successfully predicted all three classes. while tuned LightGBM improved diabetes recall and prediabetes improved from 0.00 to 2.7%. However, it remained most difficult class to detect low representation. SHAP explainability highlights the strong influence of engineered features such as HbA1c_Estimated, Glucose_Risk_Proxy, CGM_Variability_Score, BMI, and General Health. Overall, the research demonstrates feasibility of explainable multi class diabetes risk prediction and identifies key areas such as clinical data, advanced imbalance handling and transformer-based models for further future improvement.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Sahni, Anu UNSPECIFIED
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence R Medicine > Diseases > Endocrine glands - Diseases > Diabetes R Medicine > Healthcare Industry Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning R Medicine > RA Public aspects of medicine > Public Health System
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	07 Sep 2026 08:51
Last Modified:	07 Sep 2026 10:37
URI:	https://norma.ncirl.ie/id/eprint/9846

Actions (login required)

View Item