Training Stress, Sleep Behaviour, and Recovery Self-Regulation in Elite Female Boxers: A Behavioural Pattern Analysis Using Data-Driven Methods

Worrawit Rattanasateankij

Graduate School, Srinakharinwirot University, Bangkok, Thailand

Tanormsak Senakham

Department of Sports Science, Faculty of Physical Education, Sports and Health, Srinakharinwirot University, Thailand

Sirichet Punthipayanon

Department of Sports Science, Faculty of Physical Education, Sports and Health, Srinakharinwirot University, Thailan & Thailand Association of Mixed Martial Arts, Bangkok, Thailand University, Thailand.

Thidanuch Puttasimma

Department of Computer Science, Faculty of Science and Technology, Nakhon Ratchasima Rajabhat University, Nakhon Ratchasima, Thailand.

Nutcharee Senakham

Department of Sports Science, Faculty of Physical Education, Sports and Health, Srinakharinwirot University, Thailand.

Abstract

Elite boxing performance is significantly affected by training load, sleep habits, and the regulation of recovery; however, traditional monitoring approaches tend to assess these components separately and largely depend on subjective observations from coaches. Consequently, evaluating athlete readiness solely through movement analysis is insufficient, restricting both performance enhancement and the early detection of fatigue. This study investigates behavioural patterns indicative of readiness in elite female boxers and develops a data-driven classification framework that integrates physiological and behavioural metrics with boxing action execution. Data were obtained from competitive female boxers during the pre-competition phase. Training load was quantified, sleep was assessed using wearable-derived duration and efficiency measures, and recovery regulation was evaluated through standardised recovery questionnaires. Psychological dimensions, including competitive anxiety, self-confidence, team cohesion, and perception of opponents, were also recorded. A dataset containing 5,200 labelled boxing actions underwent pre-processing involving Gaussian noise filtering and feature normalisation. Principal Component Analysis (PCA) was employed to reduce data dimensionality while preserving critical structures. To capture behavioural patterns, a Dynamic Monarch Butterfly Optimized Attention-Vision AutoEncoder (DMBO-Att-VAE) was introduced, integrating behavioural and spatiotemporal motion features, followed by SoftMax classification with a train–test split. The DMBO-Att-VAE framework, implemented in Python, successfully fused behavioural and psychological metrics with spatiotemporal features, achieving a classification accuracy of 98.76% and an F1-score of 97.52%. This data-driven approach effectively identifies behavioural indicators of readiness, providing an interpretable tool to monitor performance and optimise training strategies in elite female boxing.

Behavioural Pattern Analysis, Boxing Performance, Training Stress, Sleep Behaviour, Recovery Self-Regulation, Action Recognition, Athlete Monitoring