Gradient-boosted churn model with calibrated scores
Telco churn · 7,043 customers
METHOD
LightGBM with isotonic calibration and stratified 5-fold cross-validation.
- tabular
- LightGBM
- calibration
PORTFOLIO BUILDER FOR DATA & ML
Holdout turns your experiments into a portfolio: the dataset, the method, and the metric that moved against its baseline. Build it in your browser, then download one HTML file you can host anywhere.
Free · No account · Works offline · Your data stays in this browser
Telco churn · 7,043 customers
METHOD
LightGBM with isotonic calibration and stratified 5-fold cross-validation.
HOW IT WORKS
Add a card with the dataset, the method and the outcome. Type each metric with its baseline, and say whether higher or lower is better.
Add your tools, talks, papers, jobs and degrees. Reorder or hide sections, and pick one of three themes in the live preview.
Download one HTML file with your charts inlined and no scripts. Host it on GitHub Pages, or print the résumé built from the same data.
WHAT YOU GET
| Metric | Result | vs baseline |
|---|---|---|
| Accuracy | 0.912 | +0.038 (+4.3%) improved |
| MAPE | 11.2% | −4.6 pts (−29.1%) improved |
| Weekly retrain time | 18 min | — no baseline |
Every metric can carry a baseline. Holdout works out the absolute and relative change, colours it by whether it got better, and knows that a falling error rate is good news.
Languages
ML & statistics
Group your stack into five categories, rated from Familiar to Expert, with your strongest first.
Demand forecasting service
Lab, Midnight and Journal. The preview and the export come from the same component, so what you see is what you ship.
Export a single HTML file with your chart images inlined and no JavaScript. Keep a JSON backup to move between browsers.
The same data prints as a clean, black-on-white résumé that follows your section order.
TRY IT HERE
Click Load example to open Priya’s portfolio, or start blank. Anything you type stays in this browser and is waiting when you open Holdout full-screen.
Show the result, not just the repo. Start blank An empty portfolio. Add your profile and first result card. Load example A filled-in ML engineer portfolio to explore and edit. Open HoldoutThere’s no account, no server and no analytics. Your portfolio lives in your browser’s storage until you export it, and the exported page makes no network requests.