Predictive Machine Learning Evaluates Personalized Training Pathways in Martial Arts Tricking

In a proof-of-concept study evaluating supervised machine learning for personalized sports training, an interpretable logistic regression model achieved an 81.8 percent mean accuracy in nested cross-validation and an 84.3 percent accuracy in a separate stratified hold-out evaluation when predicting trick performance.

The investigation addressed Martial Arts Tricking, an athletic discipline where diverse movement categories complicate the selection of individualized learning pathways.

To demonstrate the computational framework without drawing unwarranted inferences about real-world athlete performance, the researchers conducted their evaluations using synthetic data designed to represent plausible feature relationships rather than empirical participant measurements.

The research team evaluated multiple machine learning models to predict successful trick landings and identify influential athlete-specific features, including body weight, age, and experience. The authors selected logistic regression specifically for its interpretability, demonstrating how predicted probabilities across trick categories could highlight athlete-specific strengths and weaknesses to inform individualized learning pathway selection.

Conducted by researchers affiliated with Otto von Guericke University Magdeburg, Universitat Potsdam, and Universitat Bielefeld, the study demonstrates the feasibility of an interpretable machine-learning framework rather than a validated coaching system. The authors emphasized that real-world athlete data are required to establish practical effectiveness, while noting that the modeling approach may also transfer to other sports.

The authors emphasized that while logistic regression demonstrated strong predictive consistency on synthetic data, the current work serves strictly as a proof-of-concept machine-learning framework rather than an operational coaching system. Establishing the practical effectiveness of these predicted probabilities for individualized training pathways will require subsequent evaluation using empirical data gathered from real-world athletes.

The methodological assessment relied on structured validation protocols to measure the predictive accuracy of the logistic regression classifier. In nested cross-validation procedures, the model demonstrated an 81.8 percent mean accuracy across testing iterations. To provide an independent verification of landing classification performance, the authors also tested the classifier against a separate stratified hold-out dataset, which yielded an accuracy of 84.3 percent.

A central objective of the modeling strategy was identifying influential athlete-specific features that govern successful trick landings across diverse movement categories. The computational framework examined several key athlete attributes, including body weight, age, and experience alongside other feature variables, to assess their relative importance in predicting movement outcomes.

Rather than providing only rigid binary predictions of success or failure, the logistic regression model calculates continuous predicted probabilities across various trick categories. These probability estimates illustrate how personalized success likelihoods can pinpoint an athlete’s distinct strengths and specific weaknesses across different movement demands.

The authors explicitly characterized the current research as a preliminary proof of concept to demonstrate the computational feasibility of machine learning in personalized sports training. Because the findings were established entirely through synthetic data representing plausible feature relationships, the researchers cautioned that the model does not constitute a validated coaching system.

Subject of Research: Machine learning framework for individualized learning pathway selection in martial arts tricking

Article Title: Toward individualized learning pathway selection in martial arts tricking using predictive models

Article References: Pastel, S., Kalenov, R., Kader, H., Nübel, C., Yahiatène, S., Lorenz, N. J., & Witte, K. (2026). Toward individualized learning pathway selection in martial arts tricking using predictive models. Frontiers in Sports and Active Living, 8, Article 1963884. https://doi.org/10.3389/fspor.2026.1963884

Image Credits: AI Generated

DOI: 10.3389/fspor.2026.1963884

Keywords: martial arts tricking, supervised machine learning, logistic regression, motor learning, predictive modeling, sports coaching, performance prediction, individualized training, Toward, individualized, learning, pathway