Machine learning-based prediction and risk factor analysis of osteoporosis
Journal of Decision Making and Healthcare, Volume 2, Issue 3, December 2025, Pages: 210–223
WARAGUNT WARATAMRONGPATAI
School of Medicine, University of Phayao, Phayao 56000, Thailand
KRITTIN NARAVEJSAKUL
School of Medicine, University of Phayao, Phayao 56000, Thailand
WATCHARAPORN CHOLAMJIAK
Department of Mathematics, School of Science, University of Phayao, Phayao 56000, Thailand
PHATCHARAPON UDOMLUCK
School of Medicine, University of Phayao, Phayao 56000, Thailand
Abstract
Osteoporosis is a silent and progressive skeletal disorder characterized by reduced bone mineral density and an increased risk of fractures. Early identification of high-risk individuals is essential to enable timely interventions and prevent complications. This study aimed to develop and evaluate a machine learning–based framework for osteoporosis risk prediction by integrating demographic, medical history, and lifestyle features, with particular focus on feature reduction strategies to balance accuracy, interpretability, and efficiency. Methods: A dataset comprising 18 clinically relevant predictors was analyzed. Predictors were systematically reduced through feature selection, resulting in four distinct feature sets (Case 1–4). Five supervised learning models were implemented in MATLAB R2023b, including Fine Tree, Optimizable Support Vector Machine (SVM), Efficient Logistic Regression, Neural Network, and Optimizable Naïve Bayes. Model performance was assessed using accuracy, sensitivity, specificity, F1-score, and area under the receiver operating characteristic curve (AUC). Results: The Fine Tree model, developed using only two predictors (Age and Medications), under Case4 achieved the best balance between accuracy and computational efficiency, with 91.4% accuracy, 82.8% sensitivity, 100% specificity, an F1-score of 90.6%, and an AUC of 0.9208. The Optimizable SVM (Cases 3/4) achieved the highest predictive capability, with accuracy ranging from 90.9% to 91.4% and an F1-score of up to 90.6%, but required substantially greater computational cost. Neural Networks and Optimizable Naïve Bayes demonstrated competitive AUC values but were constrained by either sensitivity or longer training times. Conclusions: Machine learning models with minimal feature sets can achieve clinically meaningful performance in osteoporosis risk prediction. The findings suggest that streamlined models, such as Fine Tree (Case 4), are well-suited for rapid and resource-efficient screening, supporting their potential integration into real-world healthcare settings for early risk assessment and preventive care.
Cite this Article as
Waragunt Waratamrongpatai, Krittin Naravejsakul, Watcharaporn Cholamjiak and Phatcharapon Udomluck, Machine learning-based prediction and risk factor analysis of osteoporosis, Journal of Decision Making and Healthcare, 2(3), 210–223, 2025