arXiv:2512.01478v2 Announce Type: replace 
Abstract: This paper presents CourtMotion, a spatiotemporal modeling framework for analyzing and predicting game events and plays as they develop in professional basketball. Anticipating basketball events requires understanding both physical motion patterns and their semantic significance in the context of the game. Traditional approaches that use only player positions fail to capture crucial indicators such as body orientation, defensive stance, or shooting preparation motions. Our two-stage approach first processes skeletal tracking data through Graph Neural Networks to capture nuanced motion patterns, then employs a Transformer architecture with specialized attention mechanisms to model player interactions. We introduce event projection heads that explicitly connect player movements to basketball events like passes, shots, and steals, training the model to associate physical motion patterns with their tactical purposes. Experiments on NBA tracking data demonstrate significant improvements over position-only baselines: 35% reduction in trajectory prediction error compared to state-of-the-art position-based models and consistent performance gains across key basketball analytics tasks. The resulting pretrained model serves as a powerful foundation for multiple downstream tasks, with pick detection, shot taker identification, assist prediction, shot location classification, and shot type recognition demonstrating substantial improvements over existing methods.

يقدم CourtMotion إطار عمل جديدًا للنمذجة الزمانية والمكانية مصممًا لتحليل وتوقع أحداث اللعبة في كرة السلة باستخدام بيانات تتبع الهيكل العظمي. يستخدم الإطار الشبكات العصبية البيانية لالتقاط أنماط الحركة المعقدة، وهندسة Transformer لنمذجة تفاعلات اللاعبين، مما يربط في النهاية الحركات الجسدية بالأحداث التكتيكية في كرة السلة مثل التمريرات والتسديدات.

CourtMotion presenta un nuevo marco de modelado espaciotemporal diseñado para analizar y predecir eventos de juego en baloncesto utilizando datos de seguimiento esquelético. El marco emplea redes neuronales gráficas para capturar patrones de movimiento intrincados y una arquitectura Transformer para modelar las interacciones entre jugadores, conectando finalmente los movimientos físicos con eventos tácticos de baloncesto como pases y tiros.

CourtMotion présente un nouveau cadre de modélisation spatiotemporelle conçu pour analyser et prédire les événements de jeu au basketball en utilisant des données de suivi squelettique. Le cadre utilise des réseaux de neurones graphiques pour capturer des motifs de mouvement complexes et une architecture Transformer pour modéliser les interactions entre les joueurs, reliant finalement les mouvements physiques aux événements tactiques de basketball tels que les passes et les tirs.

CourtMotion introduces a new spatiotemporal modeling framework designed to analyze and predict basketball game events by utilizing skeletal tracking data. The framework employs Graph Neural Networks to capture intricate motion patterns and a Transformer architecture to model player interactions, ultimately linking physical movements to tactical basketball events such as passes and shots.

CourtMotion: Learning Event-Driven Motion Representations from Skeletal Data for Basketball

arXiv:2512.08591v1 Announce Type: new 
Abstract: Predicting the outcomes of professional basketball games, particularly in the National Basketball Association (NBA), has become increasingly important for coaching strategy, fan engagement, and sports betting. However, many existing prediction models struggle with concept drift, limited temporal context, and instability across seasons. To advance forecasting in this domain, we introduce a newly constructed longitudinal NBA dataset covering the 2004-05 to 2024-25 seasons and present a deep learning framework designed to model long-term performance trends. Our primary contribution is a Long Short-Term Memory (LSTM) architecture that leverages an extended sequence length of 9,840 games equivalent to eight full NBA seasons to capture evolving team dynamics and season-over-season dependencies. We compare this model against several traditional Machine Learning (ML) and Deep Learning (DL) baselines, including Logistic Regression, Random Forest, Multi-Layer Perceptron (MLP), and Convolutional Neural Network (CNN). The LSTM achieves the best performance across all metrics, with 72.35 accuracy, 73.15 precision and 76.13 AUC-ROC. These results demonstrate the importance of long-sequence temporal modeling in basketball outcome prediction and highlight the value of our new multi-season dataset for developing robust, generalizable NBA forecasting systems.

تقدم دراسة جديدة نموذجًا من نوع Long Short-Term Memory (LSTM) مصممًا للتنبؤ بنتائج مباريات NBA باستخدام مجموعة بيانات شاملة تغطي من موسم 2004-05 إلى 2024-25. يستخدم هذا النموذج تسلسلًا واسعًا يتكون من 9,840 مباراة لالتقاط الديناميات المتغيرة للفرق والاعتماديات بين المواسم، مما يعالج التحديات التي تواجه نماذج التنبؤ التقليدية.

Un nuevo estudio presenta un modelo de Long Short-Term Memory (LSTM) diseñado para predecir los resultados de los partidos de la NBA utilizando un conjunto de datos integral que abarca desde la temporada 2004-05 hasta la 2024-25. Este modelo utiliza una secuencia extensa de 9,840 partidos para capturar de manera efectiva las dinámicas de equipo en evolución y las dependencias entre temporadas, abordando los desafíos que enfrentan los modelos de predicción tradicionales.

Une nouvelle étude présente un modèle de Long Short-Term Memory (LSTM) conçu pour prédire les résultats des matchs de la NBA en utilisant un ensemble de données complet couvrant les saisons de 2004-05 à 2024-25. Ce modèle utilise une séquence étendue de 9 840 matchs pour capturer efficacement les dynamiques d'équipe évolutives et les dépendances entre les saisons, répondant aux défis rencontrés par les modèles de prédiction traditionnels.

A new study introduces a Long Short-Term Memory (LSTM) model designed to predict NBA game outcomes using a comprehensive dataset spanning from the 2004-05 to 2024-25 seasons. This model utilizes an extensive sequence of 9,840 games to effectively capture evolving team dynamics and dependencies across seasons, addressing challenges faced by traditional prediction models.

CourtMotion: Learning Event-Driven Motion Representations from Skeletal Data for Basketball

Was this article worth reading? Share it

LucidQuery AI

SuperMotion

Interactive Avatar