A gradient-boosted regression model (like XGBoost) trained on a national online-education marketplace can learn how course price and enrollment demand respond to features such as course length, instructor rating, category and marketing spend. Instead of a single number, the model implicitly learns a surface — for any combination of two features, it predicts a price (or popularity) that bends and tilts as those features change.
In real marketplace data, price and popularity are often only weakly correlated — the highest-priced courses are rarely the most enrolled. Tree-based models like XGBoost tend to outperform simple linear regression here precisely because they capture these kinds of threshold effects and feature interactions.
A 3D scatter of synthetic marketplace courses sits inside a live regression surface: adjust course length, instructor rating, category and marketing spend to watch the model's predicted price and popularity respond in real time.
The bending mesh is a stand-in for what a gradient-boosted model like XGBoost actually learns — price and demand are non-linear functions of the input features, not a flat plane, with real interaction effects between duration and rating.
Move the length and rating sliders to slide the glowing query marker across the surface, switch category to see the baseline shift, and toggle between the price surface and the popularity surface to compare what drives each outcome.
In real marketplace data, the priciest courses are rarely the most popular — price and demand often decouple, which is exactly the kind of pattern tree-based models are good at surfacing that linear regression misses.