Author(s): MOUSSA, I., DRIDER, A., VITEL, V., SCHORTGEN, A., LEROY, P., SAULIERE, G., GOYALLON, T., FRUCHARD, B., VAZEILLES, P., REVERET, L., TOUSSAINT, J., Institution: INSTITUT NATIONAL DU SPORT ET DE L'EXPERTISE DE LA PERFORMANCE (INSEP), Country: FRANCE, Abstract-ID: 2264

INTRODUCTION:
In the realm of sports, the notion of tactics has been delineated by various researchers as the strategic maneuvers executed by athletes to adjust to dynamically shifting conditions during matches. The delicate equilibrium of competition between two players is disrupted as soon as one scores a point, inevitably prompting alterations in tactics, playing style, and the psychological disposition of the athletes [1]. Hence, the primary objective of this study was to formulate a tactical framework for wrestling matches aimed at predicting the likelihood of match outcomes based on specific scores, remaining time until the end of the match, and associated variables throughout the bout.
METHODS:
To achieve this, data on score, time, weight class, round of competition, and wrestling style were automatically annotated from 5280 bouts sourced from the international wrestling circuit. Subsequently, a multi-state Markov model was computed to ascertain the probabilities of match progression and outcome. Transition intensities were derived using the Nelson-Aalen estimator, while the Aalen-Johansen estimator facilitated the calculation of fight evolution probabilities at each moment.
RESULTS:
The developed model provides insights into the dynamics of score transitions during a match and enables the estimation of match outcomes based on current scores. All probabilities of score changes and match outcomes were computed for every feasible score differential and consolidated into a matrix. For instance, considering a 4-point disparity, irrespective of wrestling style, this advantage tends to steer the match towards victory (40%), a tendency that becomes more pronounced if the gap arises late in the match (90% in the final minute).
CONCLUSION:
This study presented a Markov chain-based approach to enhance comprehension of the tactical progression of wrestling matches. While the results show promise, there are avenues for several enhancements, many of which necessitate a more detailed dataset. Future endeavors will involve incorporating techniques associated with each scored point as well as the location on the combat surface. The current model is poised to advance towards a semi-Markovian model that is non-uniform over time [2], considering the duration spent in a state and parameterized through a priori testing of various distributions.

[1] R. Rein et D. Memmert, « Big data and tactical analysis in elite soccer: future challenges and opportunities for sports science », Springerplus, vol. 5, no 1, p. 1410, août 2016, doi: 10.1186/s40064-016-3108-2.
[2] A. Asanjarani, B. Liquet, et Y. Nazarathy, « Estimation of semi-Markov multi-state models: a comparison of the sojourn times and transition intensities approaches », The International Journal of Biostatistics, vol. 18, no 1, p. 243‑262, mai 2022, doi: 10.1515/ijb-2020-0083.