Read-only demo
No credits charged
No private job access
Cross-Sectional Classification
Hospital Readmission Risk
Which patient records are most likely to indicate 30-day readmission risk?
A concise classification walkthrough using a clean tabular dataset, readiness preview, model comparison, and a reviewable report summary.
Demo outcome
CatBoost
Best model
Selected from replayed saved metrics.
0.790
PR AUC
Primary metric for this workflow.
42 seconds
Observed run time
Demo replay based on saved job behavior.
This workspace is a guided public replay.
Training, SHAP, report generation, prediction, job deletion, and billing actions are unavailable in demo mode.
Completed
Demo ID: demo-tabular-classification
Saved-output workflow
The demo, handbook, and job page use the same saved outputs.
Each item below comes from the completed source job, so the public replay matches what users review after a real run.
Data setup
Summary Only
The example records the data roles and checks shown in the replay.
None
Analysis checks
Summary Only
The replay explains the checks expected for this type of analysis.
None
Model explanation
Summary Only
The replay describes how model findings should be interpreted.
None
Reproducibility boundary
Available
The demo is read-only and does not run training, prediction, or billing.
None
Methods appendix
Available
A concise methods note is bundled with this public example.
methods_appendix.md
Dataset and roles
Readiness preview input
- Rows
- 1200
- Features
- 12
- Target
- readmitted_30d
- Format
- CSV
- Status
- Pass
target
features
positive class
Workflow steps
How the job moved
Preview readiness
Required roles were mapped and no blocking schema issues were found.
Compare models
Logistic regression, random forest, LightGBM, and CatBoost were compared.
Review outputs
Metrics, plots, feature importance, and report summary are available.
Saved outputs
What users review
Readiness summary
readiness
Schema fit, missingness, target balance, and role mapping.
Model performance table
metrics
Primary and secondary metrics across trained models.
AI report summary
report
A narrative generated from saved demo artifacts.
Model comparison
Replayed performance table
| Model | Status | Metric | Value | Training Time |
|---|---|---|---|---|
| CatBoost | Best | PR AUC | 0.790 | 7.8s |
| LightGBM | Trained | PR AUC | 0.764 | 5.4s |
| Random Forest | Trained | PR AUC | 0.731 | 4.9s |
| Logistic Regression | Baseline | PR AUC | 0.701 | 1.1s |
Report summary
Decision-ready notes
- The dataset passed readiness checks for a first classification run.
- CatBoost had the strongest PR AUC among the compared models.
- Results should be reviewed against clinical or operational capacity before deployment.
Model comparison
CatBoost leads
The tree-boosted model produced the strongest precision-recall tradeoff.
Feature importance
Prior utilization
Recent admissions and medication count contributed most to model separation.
Threshold review
Tune before use
The final operating threshold should match the study objective and review capacity.
Live log replay
The event stream users see during a real run
Press Replay logs to stream the curated demo events.
Ready for your own data?
Use the real workflow with authentication and billing safeguards.
A real run requires sign-in, readiness preview, visible cost estimate, and confirmation before training starts.