Research
158d agoRolling Forecasts Challenge ML Rankings
A new study shows that when air‑quality models are updated monthly, the leading machine‑learning algorithm loses its edge to a simple persistence baseline. Using 2,350 daily PM10 readings, researchers found that XGBoost performs best only under static splits, while SARIMA and persistence outperform it in rolling‑origin tests.
The Signal
The findings urge caution when deploying AI for real‑time environmental monitoring worldwide.