Article: Real time performance prediction in sports using machine learning algorithms Journal: International Journal of Data Science (IJDS) 2025 Vol.10 No.7 pp.285 - 297 Abstract: Predicting athletic performance is highly complex because it depends on many interacting factors. Conventional approaches - largely statistical analyses and expert judgement - tend to offer limited accuracy. To overcome these shortcomings, this study adopts machine-learning techniques to deliver real-time performance forecasts. First, we compiled athletes' demographic information and training records, then normalised and preprocessed the data. Drawing on extensive event-level datasets, we constructed a dynamic long short-term memory (LSTM) model that continuously captures and predicts athletes' competitive states. The model accurately anticipated swimmers' breathing patterns, with the predicted traces closely matching the observed values. It also estimated fatigue levels during the 08:00-12:00 time block: blood lactate concentrations rose from 1.2 mmol L-1 to 5.0 mmol L-1, and subjective fatigue scores climbed from 2 to 8. Comprehensive experiments across multiple sports confirm the effectiveness of the proposed real-time performance-prediction framework. Inderscience Publishers - linking academia, business and industry through research

Title: Real time performance prediction in sports using machine learning algorithms

Authors: Yunpeng Jia; Jing Liang

Addresses: Department of Sports, Beijing University of Posts Telecommunications, Beijing, 100876, Beijing, China ' Hunan Technical College of Railway High-Speed, Hengyang, 421001, China

Abstract: Predicting athletic performance is highly complex because it depends on many interacting factors. Conventional approaches - largely statistical analyses and expert judgement - tend to offer limited accuracy. To overcome these shortcomings, this study adopts machine-learning techniques to deliver real-time performance forecasts. First, we compiled athletes' demographic information and training records, then normalised and preprocessed the data. Drawing on extensive event-level datasets, we constructed a dynamic long short-term memory (LSTM) model that continuously captures and predicts athletes' competitive states. The model accurately anticipated swimmers' breathing patterns, with the predicted traces closely matching the observed values. It also estimated fatigue levels during the 08:00-12:00 time block: blood lactate concentrations rose from 1.2 mmol L^-1 to 5.0 mmol L^-1, and subjective fatigue scores climbed from 2 to 8. Comprehensive experiments across multiple sports confirm the effectiveness of the proposed real-time performance-prediction framework.

Keywords: machine learning; competitive sports; real time performance prediction; LSTM; long short-term memory network; motion data analysis.

DOI: 10.1504/IJDS.2025.151196

International Journal of Data Science, 2025 Vol.10 No.7, pp.285 - 297

Received: 21 May 2025
Accepted: 01 Aug 2025
Published online: 16 Jan 2026 *

Title: Real time performance prediction in sports using machine learning algorithms

Keep up-to-date