Datasets:

mdsajjadullah
/

fl-chex-benchmark

license: cc-by-4.0
task_categories:
  - tabular-classification
language:
  - en
tags:
  - federated-learning
  - medical
  - chest-xray
  - fairness
  - healthcare
  - benchmark
  - chexpert
  - densenet
pretty_name: FL-CheX Federated Learning Benchmark
size_categories:
  - 10K<n<100K

FL-CheX: Federated Learning Benchmark for Chest Disease Classification

Author: Md. Sajjad Ullah
Affiliation: CSE, University of Asia Pacific, Bangladesh
Date: March 2026

Dataset Summary

FL-CheX is the first federated learning benchmark combining demographic, Non-IID, and scanner heterogeneity for chest disease classification research.

15 pre-partitioned FL client nodes built from CheXpert (Stanford) with real DenseNet-121 features.

Load Dataset

from datasets import load_dataset
ds = load_dataset("mdsajjadullah/fl-chex-benchmark")

Dataset Structure

5 Demographic nodes (age × gender)
5 Non-IID Hospital nodes (disease-specialized)
5 Scanner nodes (hardware quality variation)

Citation

@misc{flchex2026,
  author = {Md. Sajjad Ullah},
  title  = {FL-CheX: A Federated Learning Benchmark},
  year   = {2026}
}