Explainable machine learning for diabetes prediction
| dc.contributor.author | Hussain, Sadam | |
| dc.contributor.supervisor | Gulliver, T. Aaron | |
| dc.date.accessioned | 2026-03-10T20:20:51Z | |
| dc.date.available | 2026-03-10T20:20:51Z | |
| dc.date.issued | 2026 | |
| dc.degree.department | Department of Electrical and Computer Engineering | |
| dc.degree.level | Master of Applied Science MASc | |
| dc.description.abstract | Diabetes is a growing global health concern, contributing to significant morbidity, mortality, and long-term economic burden. Machine Learning (ML) methods are increasingly applied to diabetes prediction, however, selecting appropriate classifiers and understanding the key features driving model decisions remain essential for reliable and clinically acceptable performance. This is particularly important in healthcare settings where clinicians may have limited familiarity with ML techniques and where transparency and trust in predictive outputs are critical. This study evaluates eight ML classifiers, Logistic Regression (LR), Random Forest (RF), Gradient Boosting (GB), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), AdaBoost (AB), Decision Tree (DT) and Neural Network (NN) using a dataset of 100,000 patient records for diabetes prediction. Models are evaluated using various configurations which includes baseline training and hyperparameter optimization using RandomizedSearchCV. The global and local interpretability is examined using SHapley Additive exPlanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME) and Explain Like I’m 5 (ELI5) to identify the most influential features contributing to predictions. These findings show that ensemble based models achieve strongest predictive performance with RF and GB outperforming other evaluated classifiers. Interpretability analyses consistently highlight that Hemoglobin A1c (HbA1c), blood glucose, Body Mass Index (BMI), and age are the dominant predictive features. A final evaluation using a reduced feature set derived with the help of Explainable AI (XAI) demonstrates that strong predictive accuracy can be maintained while improving model simplicity and interpretability. This work underscores the importance of combining ML performance with transparent feature explanations in order to support trustworthy and clinically meaningful decision support systems for diabetes prediction. | |
| dc.description.scholarlevel | Graduate | |
| dc.identifier.uri | https://hdl.handle.net/1828/23441 | |
| dc.language | English | eng |
| dc.language.iso | en | |
| dc.rights | Available to the World Wide Web | |
| dc.subject | Machine learning | |
| dc.subject | Explainable AI | |
| dc.subject | Diabetes prediction | |
| dc.subject | Feature selection | |
| dc.title | Explainable machine learning for diabetes prediction | |
| dc.type | Thesis |