Article: Predicting athletes' physical condition using multimodal machine learning models Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.98 pp.53 - 75 Abstract: Accurately predicting an athlete's physical condition is key to optimising training loads and preventing sports injuries; however, monomodal data alone struggles to capture the complex interplay between physiological and motor functions. To address this challenge, this paper proposes a method for predicting physical condition based on a multimodal machine learning model. By integrating physiological signals collected from wearable sensors with biomechanical time-series data obtained from optical motion capture, the method constructs a cross-modal attention fusion mechanism to dynamically integrate heterogeneous features. Experimental results on public multimodal sports datasets demonstrate that the proposed model achieves an area under the curve of 0.94, representing an 8.2% improvement over single-modal baseline models, with an accuracy of 0.91, significantly outperforming traditional fusion methods. The study validates the critical role of multimodal information complementarity in fitness assessment and provides a more interpretable and robust technical pathway for intelligent sports monitoring. Inderscience Publishers - linking academia, business and industry through research

Title: Predicting athletes' physical condition using multimodal machine learning models

Authors: Chenlei Huang

Addresses: Wenzhou Polytechnic University, Wenzhou, 325035, China

Abstract: Accurately predicting an athlete's physical condition is key to optimising training loads and preventing sports injuries; however, monomodal data alone struggles to capture the complex interplay between physiological and motor functions. To address this challenge, this paper proposes a method for predicting physical condition based on a multimodal machine learning model. By integrating physiological signals collected from wearable sensors with biomechanical time-series data obtained from optical motion capture, the method constructs a cross-modal attention fusion mechanism to dynamically integrate heterogeneous features. Experimental results on public multimodal sports datasets demonstrate that the proposed model achieves an area under the curve of 0.94, representing an 8.2% improvement over single-modal baseline models, with an accuracy of 0.91, significantly outperforming traditional fusion methods. The study validates the critical role of multimodal information complementarity in fitness assessment and provides a more interpretable and robust technical pathway for intelligent sports monitoring.

Keywords: physical condition prediction; multimodal fusion; sports biomechanics.

DOI: 10.1504/IJICT.2026.156259

International Journal of Information and Communication Technology, 2026 Vol.27 No.98, pp.53 - 75

Received: 30 Mar 2026
Accepted: 10 May 2026
Published online: 08 Sep 2026 *

Title: Predicting athletes' physical condition using multimodal machine learning models

Keep up-to-date