Pre-emptive Diagnosis of Osteoporosis and Osteopenia using Clinical Data

Osteoporosis and osteopenia are bone diseases characterized by low bone density and structural deterioration to an extent that makes the bones brittle and prone to fractures. The World Health Organization (WHO) differentiates osteopenia as a milder form from osteoporosis as a severe form associated with bone fractures based on statistical values. The fact that this disease is often silent, meaning it is not accompanied by noticeable symptoms, typically reveals itself when a bone fracture occurs. By then, the bones have undergone years of severe deterioration in strength and structure, which is why Bone Mineral Density (BMD) testing is essential for identifying osteoporosis and osteopenia and estimating the likelihood of future fractures. BMD is determined via a bone density test that uses dual-energy X-ray absorptiometry (DXA) to quantify the amount of minerals, such as calcium and phosphorus, inside the bones. Nevertheless, this method may be costly, which limits its accessibility. Therefore, this paper investigates the pre-emptive diagnosis of osteoporosis and osteopenia using machine learning techniques, exclusively utilizing clinical data without relying on Bone Mass Density (BMD) measurements. The dataset, which consists of 240 patients, originates from the BMD camp organized by the Unani and Panchkarma Hospital in Srinagar, J&K, India. Employing three machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM), and Gradient Boosting (GBoost)— this study leverages GridSearchCV with 10-folds for model optimization. Central to the methodology is the utilization of Sequential Forward Feature Selection (SFFS) to reduce the number of features and enhance predictive accuracy. Significantly, employing the SMOTETomek approach revealed that Random Forest outpaced other methodologies, using 20 features, where it achieved an accuracy of 91.11% with precision, recall, and F1-score values of 92%, 91%, and 91% respectively.