Datasets for Ensemble-Based Label Filtering (EBLF) Validation in Clinical Monitoring

371.8 KB of data used to validate the Ensemble-Based Label Filtering (EBLF) method for resolving label noise in clinical studies. The dataset includes simulated noisy labels for the Wisconsin Breast Cancer Dataset (569 cases, 30 features) and 281 patient observations comparing mechanical ventilation distress scales. Author Guillermo Gutierrez published the data on figshare in April 2026.

Use Cases

• Benchmarking label noise correction algorithms based on the simulated noisy label sets described.
• Training ensemble classifiers (e.g., Random Forest, XGBoost) on high-confidence concordant cases as outlined in the EBLF method.
• Studying the resolution of discordant clinical measurements, such as MV-RDOS and PmusPTP pairs, using computational adjudication.

Strengths

• Includes a well-known benchmark dataset (Wisconsin Breast Cancer Dataset with 569 cases and 30 features) for validation.
• Provides concrete performance metrics for the EBLF method, such as accuracy improvements from 0.851 to 0.970 under specific noise conditions.
• Applies the method to a real-world clinical dataset of 281 observations from 45 mechanically ventilated patients.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the primary datasets is unknown, which may limit suitability assessment.
• The clinical data's scope is limited to 45 patients, which may affect generalizability.

Provenance

Source

figshare

Collection Method

Likely contains research data from a clinical study and a publicly available benchmark dataset used for methodological validation.

Time Range

The publication date is 2026, but the temporal coverage of the underlying clinical data is not specified.

Freshness

Last updated 2026-04-22 18:40:18; freshness should be verified.

Geography

Geography is not specified in the input.