NWAOHA, PRECIOUS MMASICHUKWU (2026) A WEB-BASED DIABETES RISK PREDICTION SYSTEM USING MACHINE LEARNING FOR ENUGU STATE, NIGERIA. Other thesis, GODFREY OKOYE UNIVERSITY, ENUGU.
![[img]](https://eprints.gouni.edu.ng/style/images/fileicons/text.png) |
Text
A_20Web-Based_20Diabetes_20Risk_20Prediction_20System_20Using_20Machine_20Learni.pdf Download (2MB) |
Abstract
The problem of diabetes mellitus continues to be a public health problem in Enugu State, Nigeria, because late diagnosis is common in the region due to structural factors such as lack of laboratory facilities, expensive testing, and non-availability of self-service risk screening instruments, with a lot of cases being diagnosed at later stages when complications have already set in. This research project, therefore, aimed to create a web-based diabetes risk predictor that allows people in Enugu State to determine their individual risk of having diabetes using health and lifestyle data that do not require laboratory tests and a visit to a hospital. Logistic regression, decision tree, support vector machine, and random forest, four supervised machine learning classifiers, were used in this work and tested against each other using Pima Indians Diabetes Database (768 instances) as the main training data and 40 patient records (anonymized) obtained from UNTH, Ituku-Ozalla, and Parklane as the supplementary training data after preprocessing which involved median imputation for missing clinical data, z-score normalization, and stratified 80/20 training/testing partitioning. The model was implemented in a Flask based web app where users enter seven parameters (age, weight, height, blood pressure, family history, physical activity level and diet type), which goes through the rule-based feature estimation process to determine the clinical variables required by the model, which then categorizes the probability score as Low, Medium or High risk levels with corresponding health suggestions and a medical disclaimer. Random Forest Classifier performed the best among the models used, with an accuracy score of 79.87%, a precision of 76.47%, and an F1-Score of 68.75%. This model was chosen for deployment as a result of the above scores. All eight functional test cases worked perfectly, while the layout appeared properly on both desktop and mobile versions. These results provide proof of concept that shows that a machine learning-enabled diabetes testing web app can be developed without the use of a lab as a preventive awareness solution for people in Enugu State. It is recommended that the model be retrained using more local data in Nigeria in the future
| Item Type: | Thesis (Other) |
|---|---|
| Subjects: | H Social Sciences > H Social Sciences (General) |
| Divisions: | Faculty of Computing And Information Technology (FACIT) |
| Depositing User: | CHIBUEZE EZE |
| Date Deposited: | 04 Aug 2026 09:09 |
| Last Modified: | 04 Aug 2026 09:09 |
| URI: | http://eprints.gouni.edu.ng/id/eprint/6171 |
Actions (login required)
 |
View Item |
