The Problem
Swimming pools are frustrating when overcrowded, I wish there was a way to know how busy it'll be before you leave home. Existing "live occupancy" displays only tell you the situation right now, not in 30 or 60 minutes when you'd actually arrive.
The Solution
A continuously learning pipeline that reads sensor data every 10 minutes, trains LightGBM quantile regression models nightly, and serves probabilistic forecasts. The cone chart shows where occupancy is headed as well as how uncertain that prediction is.
How it works
12 separate models (4 time horizons × 3 quantiles) capture uncertainty that grows with the forecast horizon — shown as the widening cone. The shaded band is the 10th–90th percentile range, the line is the median forecast.