A Data-Driven University Wellness Program for Cardiovascular Disease Prevention Using Machine Learning-Based Health Risk Classification

Abstract

Annual physical examinations generate valuable employee health data; however, these records are often underutilized for preventive healthcare planning in higher education institutions. This study analyzed the health records of 478 employees of Notre Dame of Dadiangas University (NDDU), General Santos City, using machine learning classification techniques, namely ID3 entropy-based Decision Tree (DT), K-Nearest Neighbor (KNN), and Naïve Bayes (NB). The analysis focused on four cardiovascular disease (CVD) risk markers: cholesterol, creatinine, uric acid, and urinalysis. The findings revealed that 12.6% of employees had elevated cholesterol levels, 13.8% had high uric acid levels, 3.6% showed elevated creatinine levels, and 12.8% had abnormal urinalysis results. These results indicate the presence of cardiovascular and metabolic health risks among a notable portion of the university workforce. While the Decision Tree and Naïve Bayes models achieved perfect classification scores, these results are interpreted with caution due to the pre-classified nature of the dataset and the possibility of data leakage. In contrast, the K-Nearest Neighbor model demonstrated more realistic performance, with F1 scores ranging from 72% to 92% across the different health markers. Based on the identified health risks, the study developed the Marist Workplace Wellness Program for Cardiovascular Disease Prevention. The proposed program includes health education, regular screening, nutrition counseling, stress management, and preventive health interventions tailored to the needs of university employees. The study highlights the value of institutional health records in supporting evidence-based wellness initiatives and demonstrates how machine learning can assist organizations in identifying employee health risks. However, machine learning models should be used only as screening and decision-support tools and not as substitutes for professional medical diagnosis and clinical evaluation.