Enhancing Diabetes Mellitus Diagnosis and Management in Ghana through Supervised Learning: A Predictive Modeling Approach

Abstract

Purpose: The purpose of this study was to develop a supervised learning-based predictive model for the effective diagnosis and management of diabetes mellitus in Ghana, with a specific focus on how accurately can supervised learning algorithms predict diabetes in the Ghanaian context, and how can such predictions enhance healthcare decision-making?

Methods: A dataset comprising nine clinical attributes was collected from St. Michael’s Hospital, Jachie-Pramso, Ghana, containing 3,407 instances (2,232 diabetic and 1,175 non-diabetic patients). Data preprocessing involved cleansing using openRefine, normalization of binary attributes, and balancing features to reduce overfitting. Three supervised learning algorithms, Random Forest, Naive Bayes, and XGBoost, were trained and evaluated, using an 80:20 training-test split.

Results: XGBoost demonstrated the highest accuracy at 99.50%, outperforming Random Forest (98.90%) and Naive Bayes (94.50%). XGBoost also exhibited superior precision, recall, and F1-scores, highlighting its robustness in distinguishing diabetic from non-diabetic patients.

Conclusion: The study concluded that predictive modeling using supervised learning, particularly the XGBoost algorithm, offers significant potential for improving diabetes diagnosis and management in resource-constrained healthcare settings in Ghana. Implementing these predictive models into clinical decision support systems can enhance healthcare outcomes by providing accurate and timely diagnoses.