Population-Robust Radiofrequency Fracture Detection Across 14 Demographic Cohorts: A Computational Electromagnetics Study With Federated Learning and Fairness Analysis

Public data and analysis package for the manuscript: "Population-Robust Radiofrequency Fracture Detection Across 14 Demographic Cohorts: A Computational Electromagnetics Study with Federated Learning and Fairness Analysis" This repository contains:- paper2_features.mat: Feature matrix (19,488 x 156) with binary labels and feature names.- paper2_manifest.csv: Sample-level metadata (19,488 rows x 18 columns) covering ground-truth labels, demographic descriptors, and acquisition parameters.- Paper2_Complete_Analysis_Pipeline.m: MATLAB driver that reproduces the classical machine-learning, deep-learning, leave-one-group-out, fairness, and federated-learning results reported in the manuscript.- table1_dataset.csv ... table5_summary.csv: Exported manuscript tables.- README.md: Documentation and step-by-step reproduction instructions.- LICENCE.txt and CITATION.cff: Licence terms and machine-readable citation metadata. The dataset comprises 19,488 samples across 14 demographic cohorts (age x sex x BMI x bone density), generated by 2D finite-difference time-domain (FDTD) electromagnetic simulation at 2.4 GHz (with 1.8 and 3.6 GHz subsets). All experiments use a fixed random seed of 42 and are reproducible to within +/- 0.001 in AUC on the same MATLAB version. Requires MATLAB R2023b or later with the Statistics and Machine Learning Toolbox and the Deep Learning Toolbox. The 2D-FDTD simulation code that generated the raw signals is not redistributed.