DyWIS: Dynamic Weighted Influence Score-Based Machine Learning for In-Game Win Loss Prediction in Elite Badminton Singles
Badminton is a sport in which agility, anticipation, and tactical decision-making strongly influence competitive outcomes. Although recent work has shown that badminton performance and outcomes can be quantitatively modeled at the tournament and ranking levels, there remains limited research on rall...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | en |
| Published: |
Institute of Electrical and Electronics Engineers Inc.
2026
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/124764/1/124764.pdf http://psasir.upm.edu.my/id/eprint/124764/ https://ieeexplore.ieee.org/document/11460105/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Badminton is a sport in which agility, anticipation, and tactical decision-making strongly influence competitive outcomes. Although recent work has shown that badminton performance and outcomes can be quantitatively modeled at the tournament and ranking levels, there remains limited research on rally-level modeling of in-game dynamics and real-time win probability prediction. Most existing studies rely on video-derived or biomechanical features, which are computationally expensive and unsuitable for real-time applications. This study addresses these gaps by introducing a interpretable and data-driven framework that supports lightweight real-time deployment that leverages rally-level score-difference dynamics to predict in-game outcomes for elite men's singles matches. A large-scale dataset comprising 18,742 games from the Badminton World Federation (BWF) tournaments between 2018 and 2023 was collected and refined to 1,021 games involving the top 10 ranked players. We propose a novel Dynamic Weighted Influence Score (DyWIS) framework that integrates machine learning classifiers with player-specific Absolute Win (AW) and Absolute Loss (AL) thresholds, and dynamically adjusts prediction confidence using historical win rates and current rally score progression. Experiments evaluated prediction performance at multiple rally checkpoints (r = 20$ to 40), demonstrating that DyWIS provides informative early predictions with mean accuracy of above 84% at r = 20$ for several top-performing classifiers, and approaches near-deterministic accuracy at later stages as score differences enter decisive AW/AL regions. At r = 40$ , DyWIS-KNN and DyWIS-linSVM achieved the highest mean accuracies of 93.37% and 93.19%, respectively, followed by DyWIS-XGBoost at 92.85%, consistently outperforming reimplemented baselines from prior work. Overall, the results indicate that DyWIS enables earlier and more stable win/loss detection as rallies progress, supporting real-time analytics, coaching decision support, and tactical visualization in professional badminton and other fast-paced individual sports. |
|---|