The field of sports dance is highly competitive, physically demanding, and mentally challenging, leading to a high number of injuries among dancers. In order to identify the causes of these injuries, measurements were conducted on 259 dancers of various dance styles (hip-hop, rock and roll, standard and Latin American dances, and breakdance) at the Faculty of Sport. The measurements included body composition, certain motor abilities, and questionnaires to assess the occurrence of injuries, load, and basic demographic data of each participant. Some measurements were repeated twice (three months after the initial measurements). As the most common injuries in dancers are related to the ankle, knee, spine, and shoulder, predictive models for forecasting injuries in these body parts were built using various machine learning algorithms. Based on the differences between the first and second measurements, models were constructed to evaluate dancers' progress. The ReliefF algorithm was used to select the most important attributes, and the models were appropriately interpreted using the SHAP library. Logistic regression, Naive Bayes, neural networks, Support Vector Machines (SVM), Random Forests, Gradient Boosting (GB), eXtreme Gradient Boosting (XGB), and K-Nearest Neighbors (KNN) were used for modeling. For predicting knee injuries, a classification accuracy of 69 % was achieved, for spine injuries 78 %, ankle injuries 71 %, and shoulder injuries 88 %. Models for progress prediction achieved an accuracy of 98 %, enabling their practical application and identification of key factors contributing to progress.
|
