Basal spatial–temporal graph neural network for basketball tactical intent recognition and counter-path prediction

With the refinement of basketball game data analysis and the increasing demand for intelligent training, the identification of tactical intentions and the prediction of opposing strategies have become core technical support for improving teams' competitive performance. Addressing the limitations of traditional methods, including reliance on manual feature extraction, neglect of players' dynamic spatio-temporal correlations, and insufficient prediction accuracy, this paper proposes an improved spatio-temporal graph neural network model integrated with an adversarial perception module. The model first constructs a dynamic topological graph based on the spatio-temporal coordinates of players and the ball, captures collaborative dependencies among players via spatial graph convolution, and extracts temporal evolution features using gated temporal convolution. It then introduces an adversarial intensity quantification mechanism and an attention weighting module to enhance feature representation in high-confrontation scenarios, and designs a dual-task joint optimization framework to simultaneously perform tactical intention classification and adversarial path prediction. Experimental validation is conducted on a total of 7000 tactical segments from two datasets (5000 from the public Basketball Behavior Dataset (BBD) and 2000 from the self-built Chinese Professional League Dataset (CPLD)) via 5-fold cross-validation. Results show that the proposed model achieves 95.3% overall tactical intention recognition accuracy and 0.87 m mean absolute error (MAE) in path prediction on the BBD dataset, outperforming traditional machine learning and baseline deep learning models. In high-intensity adversarial scenarios, the model maintains 92.1% recognition accuracy, with 95% confidence intervals of ±1.2% for accuracy and ±0.08 m for MAE across cross-validation folds. Inference latency is 28 ms on high-end hardware (NVIDIA RTX 4090) and 87 ms on low-end deployment hardware (NVIDIA GTX 1650), which can be further reduced to approximately 50 ms via TensorRT quantization optimization, meeting basic real-time analysis requirements. The model delivers stable performance across different game quarters and tactical types, showing promising application potential for complex game scenarios. This research provides a technical reference for basketball tactical analysis, intelligent training assistance, and in-game decision support, with potential applicability in professional team training, live event analysis, and sports AI system development.

Authors

Institutions

Publication Details

Journal
Discover Artificial Intelligence
Published
2026-11-17
DOI
https://doi.org/10.1007/s44163-026-02570-9
Primary Topic
Sports Performance and Training
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Basal spatial–temporal graph neural network for basketball tactical intent recognition and counter-path prediction

Le Wu, Xingchao Xie, Kunzong Tian, Yongming Wang
4 links
Discover Artificial Intelligence
Sports Performance and Training
article

Basal spatial–temporal graph neural network for basketball tactical intent recognition and counter-path prediction

Le Wu, Xingchao Xie, Kunzong Tian, Yongming Wang
article en

Abstract

With the refinement of basketball game data analysis and the increasing demand for intelligent training, the identification of tactical intentions and the prediction of opposing strategies have become core technical support for improving teams' competitive performance. Addressing the limitations of traditional methods, including reliance on manual feature extraction, neglect of players' dynamic spatio-temporal correlations, and insufficient prediction accuracy, this paper proposes an improved spatio-temporal graph neural network model integrated with an adversarial perception module. The model first constructs a dynamic topological graph based on the spatio-temporal coordinates of players and the ball, captures collaborative dependencies among players via spatial graph convolution, and extracts temporal evolution features using gated temporal convolution. It then introduces an adversarial intensity quantification mechanism and an attention weighting module to enhance feature representation in high-confrontation scenarios, and designs a dual-task joint optimization framework to simultaneously perform tactical intention classification and adversarial path prediction. Experimental validation is conducted on a total of 7000 tactical segments from two datasets (5000 from the public Basketball Behavior Dataset (BBD) and 2000 from the self-built Chinese Professional League Dataset (CPLD)) via 5-fold cross-validation. Results show that the proposed model achieves 95.3% overall tactical intention recognition accuracy and 0.87 m mean absolute error (MAE) in path prediction on the BBD dataset, outperforming traditional machine learning and baseline deep learning models. In high-intensity adversarial scenarios, the model maintains 92.1% recognition accuracy, with 95% confidence intervals of ±1.2% for accuracy and ±0.08 m for MAE across cross-validation folds. Inference latency is 28 ms on high-end hardware (NVIDIA RTX 4090) and 87 ms on low-end deployment hardware (NVIDIA GTX 1650), which can be further reduced to approximately 50 ms via TensorRT quantization optimization, meeting basic real-time analysis requirements. The model delivers stable performance across different game quarters and tactical types, showing promising application potential for complex game scenarios. This research provides a technical reference for basketball tactical analysis, intelligent training assistance, and in-game decision support, with potential applicability in professional team training, live event analysis, and sports AI system development.

Discover Artificial IntelligenceVol. 6(1)
People’s Hospital of Rizhao (CN), Shandong Institute of Petroleum and Chemical Technology (CN)
Sports Performance and Training
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon