Determining the Level of Reaction to a Moving Object in Combat Sports Athletes Using a Composite Index and Machine...

Background. Reaction to a moving object (RMO) is an important psychophysiological function in combat sports, supporting situational perception, anticipation of an opponent's movements, and timely technical and tactical actions. Objectives. This study aimed to assess this function in combat sports athletes using a composite index and to develop a machine learning model for automated classification. Methods. The study included 312 combat sports athletes aged 10-25 years, approximately 10% of whom were female, representing five combat-sport disciplines and a broad range of sports qualifications from beginners to highly qualified athletes. Reaction to a moving object was assessed using the specialized "Reaction RMO Pro" test. The RMO_Index was calculated from standardized measures of mean absolute temporal deviation and its variability, with higher values indicating better overall performance. The model was trained using stage-specific test indicators and characteristics of premature, delayed, and accurate responses. Results. Mean absolute temporal deviation was 32.44 ± 10.29 ms, and variability was 21.88 ± 6.87 ms. The proportions of premature, delayed, and accurate responses were 48.18 ± 10.18%, 44.66 ± 10.45%, and 7.16 ± 4.40%, respectively. A balanced target variable, RMO_Level_Target, was derived from the RMO_Index: High, 104; Medium, 104; and Low, 104. Within the training-validation dataset, Create ML reported 100% classification accuracy during model development, while the model achieved 91.67% accuracy on the held-out test subset. The F1-scores were 0.93, 0.94, and 0.88 for the High, Low, and Medium classes, respectively. Conclusions. The RMO_Index provided a standardized composite measure of response timing accuracy and stability, and enabled the formation of sample-derived High, Medium, and Low performance levels. The Boosted Tree Classifier reproduced these reference levels with 91.67% accuracy in the held-out test subset. These findings support the feasibility of automated classification of RMO test results within the present sample; however, external validation is required before broader practical application.