Multimodal fusion for athlete state prediction leveraging XLNet and deep generative models

The accurate prediction of athletes’ psychological and physiological states is essential for optimizing training performance . However, current methods often struggle to effectively integrate multimodal data, limiting prediction accuracy and practical application. To address these challenges, we pro...

Full description

Saved in:
Bibliographic Details
Main Authors: Yafeng Feng, Yong Sun, Chengfang Hang
Format: Article
Language:English
Published: Elsevier 2025-10-01
Series:Alexandria Engineering Journal
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S1110016825008440
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849243035484815360
author Yafeng Feng
Yong Sun
Chengfang Hang
author_facet Yafeng Feng
Yong Sun
Chengfang Hang
author_sort Yafeng Feng
collection DOAJ
description The accurate prediction of athletes’ psychological and physiological states is essential for optimizing training performance . However, current methods often struggle to effectively integrate multimodal data, limiting prediction accuracy and practical application. To address these challenges, we propose a model that combines XLNet with a context window adjustment mechanism and Deep Generative VAE for psychological and physiological data integration. The model leverages advanced feature extraction techniques to capture both emotional tendencies from self-reports and sentiment analysis texts, as well as sequential physiological signals such as ECG and EDA. A multilayer perceptron and AdaBoost ensemble are employed for comprehensive feature fusion and state prediction. Experimental results show that our model achieves a significant improvement in classification accuracy, with a 12% increase in emotional state recognition compared to traditional models, and a 15% reduction in prediction error for physiological state estimation. Comparative experiments on the WESAD and DEAP datasets demonstrate superior performance over several baselines, validating the robustness and efficiency of our approach. These findings demonstrate the effectiveness of the proposed approach in real-time athlete monitoring and personalized training strategies. The results contribute to developing more effective tools for athletic training and provide valuable insights for optimizing physical performance and mental well-being.
format Article
id doaj-art-a64bd52a623f431a93f395d7a6ea26bb
institution Kabale University
issn 1110-0168
language English
publishDate 2025-10-01
publisher Elsevier
record_format Article
series Alexandria Engineering Journal
spelling doaj-art-a64bd52a623f431a93f395d7a6ea26bb2025-08-20T03:59:37ZengElsevierAlexandria Engineering Journal1110-01682025-10-0112992593610.1016/j.aej.2025.07.021Multimodal fusion for athlete state prediction leveraging XLNet and deep generative modelsYafeng Feng0Yong Sun1Chengfang Hang2College of Physical Education, Henan Normal University, Henan, 453007, ChinaPhysical education department, Shanghai Institute of Technology, Shanghai, 201418, China; Corresponding author.Zhejiang University, School of Software, Hangzhou, Zhejiang, 310058, USAThe accurate prediction of athletes’ psychological and physiological states is essential for optimizing training performance . However, current methods often struggle to effectively integrate multimodal data, limiting prediction accuracy and practical application. To address these challenges, we propose a model that combines XLNet with a context window adjustment mechanism and Deep Generative VAE for psychological and physiological data integration. The model leverages advanced feature extraction techniques to capture both emotional tendencies from self-reports and sentiment analysis texts, as well as sequential physiological signals such as ECG and EDA. A multilayer perceptron and AdaBoost ensemble are employed for comprehensive feature fusion and state prediction. Experimental results show that our model achieves a significant improvement in classification accuracy, with a 12% increase in emotional state recognition compared to traditional models, and a 15% reduction in prediction error for physiological state estimation. Comparative experiments on the WESAD and DEAP datasets demonstrate superior performance over several baselines, validating the robustness and efficiency of our approach. These findings demonstrate the effectiveness of the proposed approach in real-time athlete monitoring and personalized training strategies. The results contribute to developing more effective tools for athletic training and provide valuable insights for optimizing physical performance and mental well-being.http://www.sciencedirect.com/science/article/pii/S1110016825008440Quadrupole excitonMultimodal fusionPsychological statePhysiological stateAthlete monitoringEmotional classification
spellingShingle Yafeng Feng
Yong Sun
Chengfang Hang
Multimodal fusion for athlete state prediction leveraging XLNet and deep generative models
Alexandria Engineering Journal
Quadrupole exciton
Multimodal fusion
Psychological state
Physiological state
Athlete monitoring
Emotional classification
title Multimodal fusion for athlete state prediction leveraging XLNet and deep generative models
title_full Multimodal fusion for athlete state prediction leveraging XLNet and deep generative models
title_fullStr Multimodal fusion for athlete state prediction leveraging XLNet and deep generative models
title_full_unstemmed Multimodal fusion for athlete state prediction leveraging XLNet and deep generative models
title_short Multimodal fusion for athlete state prediction leveraging XLNet and deep generative models
title_sort multimodal fusion for athlete state prediction leveraging xlnet and deep generative models
topic Quadrupole exciton
Multimodal fusion
Psychological state
Physiological state
Athlete monitoring
Emotional classification
url http://www.sciencedirect.com/science/article/pii/S1110016825008440
work_keys_str_mv AT yafengfeng multimodalfusionforathletestatepredictionleveragingxlnetanddeepgenerativemodels
AT yongsun multimodalfusionforathletestatepredictionleveragingxlnetanddeepgenerativemodels
AT chengfanghang multimodalfusionforathletestatepredictionleveragingxlnetanddeepgenerativemodels