pinned
Running
Agents
Forecast Arena
⚡
Replay, rank and duel AI vs classical forecasts on real data
Gjxv PRO
Gjusev
AI & ML interests
None yet
Recent Activity
updated a dataset about 2 hours ago
Gjusev/pegelonline published a dataset 4 days ago
Gjusev/pegelonline posted an update 4 days ago
We ran almost 2,000 forecasts on real German data to see whether the new AI forecasting models are actually better.
Timestral, Chronos-2, TimesFM vs LightGBM, XGBoost and two simple baselines. Every model saw the same data and could not peek at the future. Everything ran on a CPU.
On electricity prices, the AI models won clearly. For every €100 the "same as yesterday" guess missed, Chronos-2 missed €68. LightGBM missed €86.
On train delays, it was a tie. Best AI model and best laptop-trained model: 0.07 minutes apart.
And when Frankfurt hit 61 minutes of average delay, every model predicted under 8. Nobody saw it coming.
You can replay every forecast, bet on the winner, then reveal what really happened:
Demo: https://huggingface.co/spaces/Gjusev/forecast-arena
Code: https://github.com/Gjusev/forecast-arena
#timeseries #forecastingOrganizations
None yet