A study on a method for clustering psychological traits and external characterization of physical performance in athletes based on a Gaussian mixture model
Abstract
This study examined whether psychological traits in competitive athletes can form reproducible probabilistic profiles without using physical-performance indicators to define cluster membership. Real-world data from 307 athletes were analyzed. Achievement motivation, self-control, psychological resilience, competition anxiety, and attention stability were used as the only inputs to a full-covariance Gaussian mixture model, while physical-fitness, recovery, movement-performance, and body-composition variables were retained solely for external characterization. After preprocessing, missing values were fully resolved and post-processing outlier rates ranged from 0.19% to 0.63%. The four-profile solution showed acceptable internal validity, with a Silhouette coefficient of 0.56, a Calinski–Harabasz index of 214.7, a Davies–Bouldin index of 0.71, and a mean posterior entropy of 0.27. Across 1,000 Bootstrap resamples, the mean adjusted Rand index was 0.82 and the mean normalized mutual information was 0.80. Independent validation in 180 athletes produced an adjusted Rand index of 0.79 and normalized mutual information of 0.78, while a 12 week reassessment of 120 athletes yielded a profile-agreement rate of 77.5% and Cohen’s κ of 0.71. The identified clusters represent probabilistic psychological profiles with explicit membership uncertainty, rather than fixed physical-performance categories or causal explanations. The model may serve as an auxiliary tool for psychological monitoring, boundary-athlete identification, and individualized athlete-support decisions.