A Machine Learning-Based Bayesian network model for predicting osteoporosis in postmenopausal Thai women

Osteoporosis causes a significant health challenge, especially among postmenopausal women. Accurate and efficient screening methods are crucial for early detection and intervention. This study addresses the limitations of traditional screening approaches by developing a novel Bayesian Network (BN) model specifically for osteoporosis screening in postmenopausal Thai women. Data preprocessing involved imputing missing values using the 10-nearest neighbor method and applying three discretization techniques—manual, unsupervised, and supervised methods—to a dataset comprising 90 normal cases and 266 abnormal cases. The study demonstrates that the BN model, employing the supervised discretization method known as Fayyad and Irani’s Minimum Description Length (MDL), achieves the highest performance in terms of accuracy (0.772), specificity (0.284), precision (0.796), F-measure (0.860), and G-mean (0.487). The performance of the BN model was compared with benchmark models, including logistic regression (LR), support vector machines (SVM), artificial neural networks (ANN), and decision trees (DT). The results indicated comparable performance across all models. The Bayesian Network for Postmenopausal Osteoporosis (BNPOS) emerges as a sensitive and versatile tool for osteoporosis screening, with impressive performance metrics, including an accuracy of 0.789, sensitivity of 0.984, specificity of 0.218, precision of 0.787, F-measure of 0.875, and G-mean of 0.442. It uncovers intricate direct and indirect relationships between risk factors and osteoporosis, identifying age, body mass index (BMI), arm circumference, weight, and hip circumference as key predictors. The BNPOS model relies on probabilistic inference, which minimizes the impact of missing data by predicting outcomes using only age and BMI as the most critical predictors. This probability-based approach makes it particularly well-suited for use in resource-limited settings, providing valuable insights for healthcare professionals.