Prediction of Diabetes Status Using Binary Logistic Regression with Artificial Neural Network

Zaharadeen Abdullahi, Adamu Abubakar and A. U. Shelleng

Department of Mathematical Sciences, Gombe State University, Gombe, Nigeria

*Corresponding author’s
Email: abumalala@gsu.edu.ng, doi.org/10.55639/607.020100108


ABSTRACT

ABSTRACT Diabetes is among the most prevalent non communicable diseases globally, contributing significantly to morbidity, mortality, and healthcare costs, particularly in low- and middle-income countries. Early detection of this condition is essential for effective prevention and management; however, traditional diagnostic approaches are often costly, time consuming, and inaccessible to many populations. This study developed and evaluated predictive models for diabetes using Binary Logistic Regression (BLR), Artificial Neural Networks (ANN), and a weighted Ensemble model that integrates both approaches. The study uses retrospective patient data obtained from Federal Teaching Hospital Gombe and Specialist Hospital Gombe, Nigeria. A total of 3,146 patient records containing demographic, clinical, lifestyle, behavioral and environmental variables were analyzed. Data preprocessing included cleaning, encoding of categorical variables, normalization of continuous variables, and outlier handling. Model performance was evaluated using accuracy, sensitivity, specificity, precision, F1 score, and Area Under the Receiver Operating Characteristic Curve (AUC ROC). Model stability was assessed using fivefold cross validation. Results showed that ANN model achieved the highest accuracy (74.72%) among individual models, while the Ensemble model produced the best overall discrimination with an AUC ROC of 0.792 and an F1 score of 79.08% on the test set. Logistic regression models provided interpretable insights into significant risk factors but were outperformed by ANN and Ensemble models in predictive accuracy. The findings demonstrate that integrating traditional statistical methods with machine learning techniques improves predictive performance and model robustness. The Ensemble modeling approach offers a balanced and effective framework for early detection of diabetes.

KEYWORDS

Ensemble,
Binary-Logistic
Regression,
Artificial-Neural
Network.

.