Abstract Preview:
Executive Summary Background and objectivesDiabetes mellitus (DM) and hypertension (HTN) are leading causes of cardiovascular disease, death, and disability, with a growing burden in developing countries. Early detection is essential, and machine learning (ML) offers powerful tools for predicting diseases risk by uncovering complex patterns in health data. At the same time, the Health Belief Model (HBM) explains preventive behaviors through constructs such as perceived susceptibility, severity, benefits, barriers, self-efficacy, and cues to action. This study integrates ML-based predictive modeling with the HBM to identify individuals at risk of HTN and DM and to better understand the behavioral factors influencing prevention, employing a dataset collected in 2025. Materials and methodsData on DM and hypertension HTN were collected from 1,771 employees of Debre Markos, Injibara, and Bahir Dar universities in Northwest Ethiopia. The cross-sectional survey included demographic, health-related, and behavioral factors, with constructs from the Health Belief Model (HBM) such as perceived susceptibility, severity, benefits, barriers, self-efficacy, and cues to action. RFE was applied to identify the most relevant predictors of DM and HTN. Four machine learning algorithms—Logistic Regression (LR), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and k-Nearest Neighbor (kNN) were developed using theselected features. Model performance was evaluated based on accuracy, precision, recall, the F1-score, and the area under the ROC curve. ResultsThis study found that the ensemble ML models, RF and GBDT, outperformed in predicting HTN and DM, achieving higher accuracy, precision, recall, F1-score, and area under the ROC curve. Analysis of Health Belief Model (HBM) constructs further showed that preventive behaviors were positively associated with perceived susceptibility, severity, benefits, self-efficacy, and cues to action, while perceived barriers were negatively associated. Perceived susceptibility emerged as a significant predictor of HTN and DM, and cues to action contributed to the identification of undiagnosed DM cases.
Full Abstract:
Executive Summary Background and objectivesDiabetes mellitus (DM) and hypertension (HTN) are leading causes of cardiovascular disease, death, and disability, with a growing burden in developing countries. Early detection is essential, and machine learning (ML) offers powerful tools for predicting diseases risk by uncovering complex patterns in health data. At the same time, the Health Belief Model (HBM) explains preventive behaviors through constructs such as perceived susceptibility, severity, benefits, barriers, self-efficacy, and cues to action. This study integrates ML-based predictive modeling with the HBM to identify individuals at risk of HTN and DM and to better understand the behavioral factors influencing prevention, employing a dataset collected in 2025. Materials and methodsData on DM and hypertension HTN were collected from 1,771 employees of Debre Markos, Injibara, and Bahir Dar universities in Northwest Ethiopia. The cross-sectional survey included demographic, health-related, and behavioral factors, with constructs from the Health Belief Model (HBM) such as perceived susceptibility, severity, benefits, barriers, self-efficacy, and cues to action. RFE was applied to identify the most relevant predictors of DM and HTN. Four machine learning algorithms—Logistic Regression (LR), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and k-Nearest Neighbor (kNN) were developed using theselected features. Model performance was evaluated based on accuracy, precision, recall, the F1-score, and the area under the ROC curve. ResultsThis study found that the ensemble ML models, RF and GBDT, outperformed in predicting HTN and DM, achieving higher accuracy, precision, recall, F1-score, and area under the ROC curve. Analysis of Health Belief Model (HBM) constructs further showed that preventive behaviors were positively associated with perceived susceptibility, severity, benefits, self-efficacy, and cues to action, while perceived barriers were negatively associated. Perceived susceptibility emerged as a significant predictor of HTN and DM, and cues to action contributed to the identification of undiagnosed DM cases.