IHEKWABA, KACHISICHO C. (2026) DEVELOPMENT OF A HYBRID MACHINE LEARNING SYSTEM FOR SICKLE CELL DISEASE DETECTION AND CRISIS PREDICTION. Other thesis, Godfrey Okoye University, Enugu.
|
Text
DEVELOPMENT OF A HYBRID MACHINE LEARNING SYSTEM FOR SICKLE CELL DISEASE DETECTION AND CRISIS PREDICTION.pdf Download (2MB) | Preview |
Abstract
Nigeria is a nation with the most cases of Sickle Cell Disease with approximately 150,000 cases born annually, yet existing diagnostic methods cannot predict crises and remain costly and equipment-dependent. This study addresses three gaps in prior Nigerian Sickle Cell Disease machine learning research; reliance on single algorithms, absence of integrated image and crisis prediction, and lack of a deployable web system. MediPredict, a hybrid Machine Learning based Sickle Cell Disease prediction system, was developed combining a MobileNetV2 CNN for blood smear image classification and a weighted Random Forest-XGBoost ensemble for clinical crisis risk prediction, deployed as a role-based web application for four actors; Administrator, Clinician, Laboratory Technician and Patient. A clinical dataset of 3,000 records was constructed from haematological parameter distributions derived from published Nigerian SCD studies including Nnodu et al. (2019), Ezenwosu et al. (2021), and Okon et al. (2024). The CNN achieved 97.43% accuracy and 99.76% AUC on 716 African blood smear images while the hybrid ensemble achieved 96.67% accuracy and 99.31% AUC a 17.67 percentage point improvement over the 79% reported by Okon et al. (2024). The system supports image upload, crisis prediction, explainable patient results, and clinician feedback, demonstrating the feasibility of an integrated AI-powered SCD management system for the Nigerian healthcare context.
| 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 09:51 |
| Last Modified: | 22 Jul 2026 09:51 |
| URI: | http://eprints.gouni.edu.ng/id/eprint/5969 |
Actions (login required)
 |
View Item |
