Abstract
OBJECTIVE: The objective of this research was to develop a machine learning-based predictive model for osteoporosis screening using demographic and clinical data, including T-scores derived from calcaneus Quantitative Ultrasound (QUS). The study aimed to offer a cost-effective and accessible alternative to Dual-Energy X-ray Absorptiometry (DXA) scans, especially in resource-constrained settings. RESULTS: The model achieved a classification accuracy of 88%, outperforming traditional decision trees by 10%. This improvement in accuracy demonstrates the potential of the random forest algorithm in identifying patients at risk of osteoporosis. Misclassification rates were minimal, with most errors occurring in distinguishing osteopenia from normal cases. The findings indicate that machine learning models trained on QUS data can aid in early identification of osteoporosis, reducing reliance on costly DXA scans.