Abstract
The combination of sports psychology and new wearable technology is allowing experts to assess psychological and cognitive performance in elite basketball more accurately. This study investigates the application of Human Activity Recognition (HAR) using wrist-worn inertial measurement unit (IMU) sensors to infer psychological resilience and decision-making quality in basketball players. We utilized the Hang-Time HAR dataset comprising 24 participants across structured drills and unstructured game scenarios, extracting temporal, spatial, and frequency-domain features from accelerometer data. Machine learning models, including Support Vector Machines (SVM), Random Forest (RF), and Gradient Boosting, were employed to classify psychological states. Findings indicated that resilience and decision-making accuracies under stratified five-fold cross-validation were 84.7% (95% CI: 81.2788.1) and 78.3% (95% CI: 74.6782.0), respectively. To mitigate the risk of data leakage, a leave-one-subject-out validation was conducted, resulting in high accuracies of 79.4 and 73.1, respectively. The significance of the observed differences was confirmed by statistical analyses (post-hoc tests, effect sizes). In comparison with the current research in HAR and basketball, which is mainly dedicated to the technical capabilities or activity recognition, this study is the first one to combine IMU-based sensing with psychological state inference. The results demonstrate the potential of wearable movement analytics to provide real-time, ecologically valid measurements of cognitive-psychological variables in sport, to inform the monitoring of players, the customization of training, and the study of sports psychology.