RESEARCH · RESEARCH · #2708
Learning sports time-to-arrival via quantile regression; MLP outperforms physics-based models
The arXiv preprint casts time-to-arrival (TTA) — the shortest plausible time for a player to reach a location — as a conditional quantile regression problem and trains a lightweight multilayer perceptron (MLP) on optical tracking data from all 64 matches in the PFF FC FIFA World Cup 2022 dataset. The MLP is evaluated against six physics-based player-motion models re-expressed as TTA predictors and a learned reachable-hull baseline, consistently outperforming those baselines; the authors also probe transferability by testing on basketball tracking data and a single-game NFL tracking dataset and publish code at mguti97/TTA-Quantile.
KEY POINTS
- The arXiv preprint casts time-to-arrival (TTA) — the shortest plausible time for a player to reach a location — as a conditional quantile regression problem and trains a lightweight multilayer perceptron (MLP) on optical tracking data from all 64 matches in the PFF FC FIFA World Cup 2022 dataset.
- The MLP is evaluated against six physics-based player-motion models re-expressed as TTA predictors and a learned reachable-hull baseline, consistently outperforming those baselines; the authors also probe transferability by testing on basketball tracking data and a single-game NFL tracking dataset and publish code at mguti97/TTA-Quantile.
- A learned quantile-based TTA estimator that outperforms physics-based baselines can give more accurate, data-driven measures of player reachability and space control and may generalize across sports analytics tasks.
WHY IT MATTERS
A learned quantile-based TTA estimator that outperforms physics-based baselines can give more accurate, data-driven measures of player reachability and space control and may generalize across sports analytics tasks.