NNANNA, SAMUEL OGBONNA (2026) DESIGN OF AN EXPLAINABLE MOBILE HEALTH DECISION-SUPPORT SYSTEM FOR DIABETES RISK ASSESSMENT USING A HYBRID DEEP LEARNING APPROACH. Other thesis, Godfrey Okoye University, Enugu.
|
Text
NNANNA SAMUEL OGBONNA(1).pdf Download (2MB) | Preview |
Abstract
Diabetes mellitus remains a major global health challenge, with millions of cases undiagnosed, particularly in sub-Saharan Africa. Existing risk prediction systems are largely limited by isolated feature consideration, insufficient integration of genetic and lifestyle data, and a lack of explainability. This study designed and developed an explainable mobile health decision-support system for diabetes risk assessment using a hybrid deep learning approach. The methodology adopted a design science research framework, leveraging the CDC BRFSS Diabetes Health Indicators dataset containing over 70,692 records and 22 features. The proposed hybrid CNN-Transformer architecture was implemented and trained alongside baseline models including Logistic Regression, Random Forest, and XGBoost for comparative evaluation. The hybrid model integrated convolutional layers for local feature extraction with a Multi-Head Attention Transformer block to capture contextual relationships among healthcare indicators, and was optimised using residual connections, layer normalisation, dropout, and early stopping. SHAP explainability techniques were integrated to provide transparent and interpretable prediction outputs. Experimental results demonstrated competitive performance, with XGBoost achieving the highest testing accuracy of 75.52%, while the proposed CNN-Transformer model achieved 74.89% with superior explainability, scalability, and contextual feature learning. The SHAP analysis identified General Health, BMI, High Blood Pressure, Age, and High Cholesterol as the most influential predictors. A mobile-based decision-support interface was also designed and implemented to support real-time diabetes risk assessment. The study demonstrates the feasibility of integrating hybrid deep learning and explainable AI within a mobile health framework for early diabetes detection and preventive healthcare.
| Item Type: | Thesis (Other) |
|---|---|
| Subjects: | Q Science > Q Science (General) |
| Divisions: | Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science |
| Depositing User: | MICHAEL MADUBUKO |
| Date Deposited: | 22 Jul 2026 11:46 |
| Last Modified: | 22 Jul 2026 11:46 |
| URI: | http://eprints.gouni.edu.ng/id/eprint/6002 |
Actions (login required)
 |
View Item |
