Neural Networks for NBA Player Movement Forecasting
A new arXiv paper (2605.14855) evaluates machine learning models for forecasting dynamic movements of NBA players, a task complicated by abrupt velocity and direction changes. Traditional methods like (S)ARIMA(X), Kalman filters, and Particle filters struggle with non-linear dynamics. ML approaches such as LSTM networks, graph neural networks (GNNs), and Transformers offer better accuracy but fail to capture temporal dependencies and contextual interactions in chaotic sports environments. The study assesses these models' strengths and weaknesses, with experimental results highlighting key performance insights.
Key facts
- arXiv paper 2605.14855 evaluates ML models for NBA player movement forecasting.
- Traditional methods (S)ARIMA(X), Kalman filters, Particle filters struggle with non-linear dynamics.
- ML methods LSTM, GNNs, Transformers offer greater accuracy but miss temporal-contextual interplay.
- Abrupt velocity and direction changes in sports pose forecasting challenges.
- Paper assesses strengths and weaknesses of evaluated models.
- Experimental results reveal key performance insights.
- Forecasting is crucial for mitigating delays in signal processing pipelines.
- Study focuses on dynamic movements of NBA players.
Entities
Institutions
- arXiv