What you learn
How to clean the data, engineer features, evaluate by participant grouping, and interpret a baseline with clear limits.
Teaching Case
Kaggle Teaching Case
This is a workflow-learning page for baseline construction, not a page of strong clinical or wearable evidence.
Who it helps
AI / DS newcomers, developers building quick prototypes, and medical readers who want to see what a data workflow looks like end to end.
How not to misread it
It is not a wearable study, not a clinically anchored label set, and not enough evidence on its own for production product claims.
Case overview
| Item | Value |
|---|---|
| Raw file | FedCycleData071012 (2).csv |
| Clean modelling table | 1,660 rows × 12 columns |
| Main target | LengthofCycle |
| Evaluation strategy | GroupKFold / participant-level grouping |
| Analysis code | The local notebook is not bundled with the current website repo |
Three things this case teaches well
Data preparation often matters more than model choice
Variable filtering, missing-data decisions, and feature definitions shape the result before architecture does.
GroupKFold is not a side detail
If the same participant leaks across train and test, the result will be systematically inflated.
Public tabular data are good for baselines, not overclaiming
This is a useful sandbox for experiment design, but it is not a replacement for stronger labels or multimodal physiology.
Linked files
This repo currently keeps the teaching page only. The original notebook, raw CSV, cleaned CSV, and supporting documentation are not bundled in this website snapshot.
Analysis notebook
Not bundled in the current website repo.
Dataset documentation
Not bundled in the current website repo.
Raw CSV
Not bundled in the current website repo.
Clean CSV
Not bundled in the current website repo.