Automated evaluation of basketball skills using multimodal data fusion for improving shooting training efficiency

Abstract

Reliable basketball shooting assessment requires integration of complementary movement signals and actionable feedback. This study developed the multimodal shooting evaluation and efficiency model (MSEEM) and evaluated it in 156 players assigned to MSEEM feedback or standard coaching. RGB video, pose, inertial measurement unit signals, ball trajectories, and shot outcomes were fused to estimate rubric-based total and component scores. MSEEM achieved lower mean absolute error (4.82 ± 0.31) and stronger rank agreement ( ρ = 0.824) than SlowFast (6.24 ± 0.41; ρ = 0.702). MSEEM feedback also produced steeper training improvement and higher 14-day retention ( d = 0.68), supporting standardized component-targeted feedback.