Estimate a quality property between laboratory assays without using future truth, while showing when the model is out of domain and should abstain.
3 · What it produces
Estimate/split-conformal interval, domain status, OOD check and frozen-benchmark context.
The images come from demonstration runs included in the attached repository and are shown without dates to illustrate workflow structure, method and evidence.
4 · How it works
From data to output, with explicit controls.
1Synthetic process sensors
2Delayed laboratory with noise/QC
3Linear/PLS/Ridge/RF/GB comparison
4Split-conformal calibration
5Isolation Forest OOD
6Abstention and risk-coverage diagnostics
Transferable pattern: Delayed-label estimation + OOD abstention.
5 · Data, AI/ML/RPA and method
Data and features
process sensors + delayed laboratory with noise/QC
Calculation / inference
Linear, PLS, Ridge, Random Forest, Gradient Boosting; Isolation Forest OOD
Validation
RMSE/MAE/bias/coverage, skill versus last lab, split-conformal and truth-vs-lab; risk-coverage diagnostic only
Stack
Python, pandas, scikit-learn, SQLite, Matplotlib
6 · Technical dossier
Operating conditions and declared limitations.
| Compute cadence | Every 6 h: 01:50/07:50/13:50/19:50 UTC |
|---|---|
| Planned publication | IT Sat 10:30; EN Sat 16:30 |
| Possible applications | soft sensing, inferential measurement and quality monitoring |
| Research use | conformal coverage, OOD as domain distance and model error versus laboratory uncertainty |
| Required condition | truth and laboratory are simulated; no real performance claim |
| Failure mode | leakage from true values or a soft sensor answering outside its domain |
Interpretation boundary
Process, truth and laboratory are simulated. Coverage and skill describe the stated benchmark; real use requires plant data, Gage R&R/QC and prospective validation.
7 · Verification and further reading