README.md

metadata

license: other
tags:
  - missing-data
  - imputation
  - benchmark
  - healthcare

Missing-Data Prediction Benchmark — results

Two experiments on medical datasets under simulated missingness (MCAR/MAR/MNAR, rates 0.1–0.5, seeds 42/1/7), reported as mean [95% CI].

This release contains AFA-engine acquisition results (curves, CI tables, examples, AUROC, and per-engine token accounting) and cached question banks at a uniform missing rate of 0.3 across five classification datasets (arrhythmia, breast_cancer, gallstone, heart_disease, parkinsons) and four regression datasets (diabetes, nhanes, who_life_expectancy, medical_cost).

Scenario 1 and the Scenario 1 vs Scenario 2 A/B comparison are not refreshed by this update. The staged raw-acquisition tree intentionally contains only Scenario 2; an incremental Scenario-2 upload preserves the previously released remote Scenario-1 artifacts. The per-engine Scenario-2 token table uses all Scenario-2 samples; matched-sample token comparison will follow with the complete Scenario-1 release. MIMIC-III is unsupported by the tabular AFA runner (expected incompatibility skips).

How to read (no download needed)

Every CSV opens in HuggingFace's Dataset Viewer (sortable, in-browser):

• experiment1_prediction/prediction_classification_long.csv / prediction_regression_long.csv — one tidy row per (method, dataset, mechanism, rate, seed) with all metrics + status. This is the clean, flat view of every run (no folder-diving).
• experiment1_prediction/classification_mean_ci.csv / regression_mean_ci.csv — the same, collapsed to mean [95% CI] over seeds (blank cells carry a Status).
• experiment2_acquisition/curves_long.csv — per-budget Scenario-2 line-plot data.
• experiment2_acquisition/acquisition_curves_ci.csv / acquisition_final_ci.csv — seed-aggregated Scenario-2 tables.
• experiment2_acquisition/acquisition_examples.csv — qualitative acquisition examples.
• experiment2_acquisition/scenario2_method_token_accounting.csv — per-engine cached and uncached Scenario-2 token accounting.
• DATA_DICTIONARY.md — column definitions and status meanings.
• raw_prediction/ — per-cell raw Exp-1 metric JSON (in-scope datasets, canonical layout, see raw_prediction/README_LATEST.md).
• raw_acquisition/ — per-cell raw Scenario-2 AFA traces and question banks (no model weights).

Data-use / MIMIC-III notice

One regression dataset (mimic_lengthofstay) is derived from MIMIC-III, credentialed under the PhysioNet Data Use Agreement. This repository contains only aggregate derived metrics — no patient-level records, no raw MIMIC data, and no models trained on MIMIC.

Generated 2026-08-28 by scripts/reporting/prepare_hf_upload.py.