Identifying Weekly Physical and Physiological Profiles in Professional Basketball Using Heart Rate Variability and Training Load Clustering

This study applied an unsupervised machine-learning technique to characterize weekly physiological-load states in professional basketball players using autonomic (HRV), cardiovascular (SHRZ), and mechanical (Movement Load) parameters. Eleven male professional players (mean = 25.5 years, SD = 3.9 years) were monitored over 18 weeks. Weekly averages of heart rate variability (LnRMSSD), internal load (SHRZ), and external load (Movement Load) were computed and standardized within players (z-scores). K-means clustering algorithm was applied to zHRV, zSHRZ, and zML, with the number of clusters selected via elbow criterion and clustering quality using the silhouette coefficient. Repeated-measures correlations were used to quantify associations among variables. Three distinct profiles were identified (p < .001), with values reported as mean (SD): (1) High load-elevated HRV (zHRV = 0.66 (0.56); zSHRZ = 0.69 (0.58); zML = 0.71 (0.56)); (2) Low load-near-average HRV (zHRV = 0.15 (0.84); zSHRZ =-0.98 (0.73); zML = -1.01 (SD = 0.65)); (3) High load-reduced HRV (zHRV = -1.04 (0.64); zSHRZ = 0.33 (0.59); zML = 0.34 (0.63)). Profiles recurred across the season, and heatmap visualization showed substantial temporal and inter-individual variability. Repeated-measures correlations showed weak HRV-SHRZ (r = -0.18, 95% CI [-0.20,-0.05], p = .007) and HRV-ML (r = -0.06, 95% CI [-0.19, -0.07], p = .368) relationships, but a very large SHRZ-ML association (r = 0.87, 95% CI [-0.84,-0.90], p < .001). Unsupervised clustering offers a practical and data-driven framework to summarizing weekly physiological-load states in elite, potentially improving individualized load-management strategies. Future studies should investigate the extent to which these clustered profiles are associated with performance outcomes, fatigue markers, and injury risk.