In silico augmentation strategies for enhanced machine learning performance in fracture recognition.

Area of Science:

• Biomedical data analysis
• Machine learning in healthcare
• Synthetic data generation

Background:

• Fracture risk prediction is crucial for patient care.
• Machine learning offers potential for improving risk assessment.
• Generating realistic synthetic biomedical data presents challenges.

Purpose of the Study:

• To develop and validate a machine learning framework for fracture risk prediction.
• To assess the utility of synthetic biomedical data for model training and validation.
• To compare the performance of various classification models for fracture risk prediction.

Main Methods:

• Analysis of a retrospective dataset (169 patients) with variables like age, sex, bone mineral density (BMD).
• Evaluation of multiple machine learning models (Logistic Regression, Random Forest, Gradient Boosting, SVM, ensemble voting).
• 5-fold stratified cross-validation and assessment of synthetic data fidelity (distribution, correlation, transferability).

Main Results:

• The Voting Hard ensemble classifier achieved 85.8% accuracy and 0.822 F1-score.
• Logistic Regression showed the highest discriminative capability with an AUC of 0.88.
• Synthetic data closely matched real data in marginal distributions but had weaker inter-feature correlation preservation.

Conclusions:

• Ensemble machine learning methods show promise for accurate fracture risk prediction.
• Rigorous validation is essential when using synthetic biomedical data.
• The developed framework supports privacy-preserving synthetic data applications in biomedical machine learning.