A novel classification method for balance differences in elite versus expert athletes based on composite multiscale complexity index and ranking forests.

Balance is crucial for various athletic tasks, and accurately assessing balance ability among elite athletes using simple and accessible measurement methods is a significant challenge in sports science. A common approach to balance assessment involves recording center of pressure (CoP) displacements using force platforms, with various indicators proposed to distinguish subtle balance differences. However, these indicators have not reached a consensus, and it remains unclear whether these analyses alone can fully explain the complex interactions of postural control. In this study, we investigated four parameters related to balance control-anterior-posterior (AP) displacement, medial-lateral (ML) displacement, length, and tilt angle-in 13 elite athletes and 12 freestyle skiing aerial expert athletes. Data were recorded during 30-second trials on both soft and hard support surfaces, with eyes open and closed. We calculated the CMCI and used four machine learning algorithms-Logistic Regression, Support Vector Machine(SVM), Naive Bayes, and Ranking Forest-to combine these features and assess each participant's balance ability. A classic train-test split method was applied, and the performance of different classifiers was evaluated using Receiver Operating Characteristic(ROC) analysis. The ROC results showed that traditional time-domain features were insufficient for accurately distinguishing athletes' balance abilities, whereas CMCI performed the best overall. Among all classifiers, the combination of CMCI and Ranking Forest yielded the best performance, with a sensitivity of 0.95 and specificity of 0.35. This nonlinear, multidimensional approach appears to be highly suitable for assessing the complexity of postural control.