Feature Impact Analysis on Top Long-Jump Performances with Quantile Random Forest and Explainable AI Techniques

Qi Gan, Stéphan Clémençon, Mounîm A. El-Yacoubi, Sao Mai Nguyen, Eric Fenaux, Ons JelassiarXiv 2025 · DOI · arXiv

Introduction

In elite sports, understanding the "average" athlete is often insufficient for high-performance coaching. Success at the world-class level is defined by outliers—the gold medalists who perform at the extreme right tail of the performance distribution. Traditional biomechanical analysis often relies on rigid physics-based models or standard "least-squares" regression, which optimizes for mean performance. This approach can lead to a "regression to the mean," potentially obscuring the specific technical nuances that drive elite results.

This study addresses this gap by employing Quantile Random Forests (QRF) to model the 90th percentile of long-jump performances. By focusing on the top 10% of jumps, the researchers aim to identify the biomechanical "secrets" that separate good jumps from world-class ones.

Dataset and Distribution

The study utilized a specialized dataset of 68 jumps (35 men, 33 women) harvested from the official biomechanical reports of three major World Championship finals: 2009 Berlin (Outdoor), 2017 London (Outdoor), and 2018 Birmingham (Indoor). The dataset includes 45 numeric features covering approach velocities, step lengths, flight times, and joint angles at various phases.

Distribution of effective distance (d_resEffe) for men and women athletes, highlighting the 90th percentile target for quantile regression.

As shown in the distribution plots, the target variable 'd_resEffe' (effective distance) exhibits clear separation between genders, and the 90% quantile represents the elite threshold the model seeks to predict.

Methodology

The research pipeline was specifically designed to handle small, high-stakes datasets through several key stages:

  1. Data Imputation: To address missing values in competition reports, the authors found that Random Forest iterative imputation was most effective for men’s data, while K-Nearest Neighbors (KNN) performed better for women’s data.
  2. Feature Selection: Lasso-regularized linear quantile regression was used to reduce the 45 biomechanical metrics to the most impactful 10–19 features. This process balances model complexity (number of features) against predictive error (pinball loss).
  3. Quantile Random Forests (QRF): Unlike standard models, QRF provides a full conditional distribution, allowing for targeted analysis of the 90th percentile.
  4. Explainable AI (XAI): Tools like SHAP (SHapley Additive exPlanations) and ICE (Individual Conditional Expectation) plots were used to interpret the "black box" model and identify non-linear thresholds.
Feature selection process using Lasso-regularized linear quantile regression for both genders, showing the trade-off between pinball loss and the number of non-zero features.

Key Insights

The analysis revealed distinct technical strategies and thresholds critical for elite performance:

  • The "Stiff Leg" Threshold: For men, a critical threshold was identified for the knee angle of the supporting leg at touchdown (a_knee_TD). Performance improved significantly when this angle exceeded 169°, suggesting that a stiffer leg better transfers horizontal momentum into vertical lift.
  • Gender-Specific Drivers: Success factors differ by gender. Men’s performance is dominated by take-off technique and velocity. For women, approach step adjustments (e.g., r_stepDiff_S21) and landing poses (a_knee_LD) play a more significant role in reaching the top tier.
  • Velocity vs. Technique: While horizontal velocity (v_H_S1) is the primary predictor for all athletes, ICE plots showed diminishing returns at extreme speeds. At the highest velocities, technical features like step length changes become the deciding factors for podium finishes.

Results and Conclusion

The QRF model achieved high predictive accuracy, with mean pinball losses of 0.0287 for men and 0.0333 for women. The study successfully demonstrated that focusing on the 90th percentile reveals biomechanical thresholds—such as specific joint angles and velocity pairings—that are invisible in mean-based models.

While the small sample size (N=68) presents limitations, this framework provides a powerful tool for sports scientists to move beyond averages and provide actionable, data-driven insights for elite athletes. Future work will look to incorporate larger datasets and more advanced XAI techniques to further refine these technical models.

Summary updated

Abstract

Biomechanical features have become important indicators for evaluating athletes' techniques. Traditionally, experts propose significant features and evaluate them using physics equations. However, the complexity of the human body and its movements makes it challenging to explicitly analyze the relationships between some features and athletes' final performance. With advancements in modern machine learning and statistics, data analytics methods have gained increasing importance in sports analytics. In this study, we leverage machine learning models to analyze expert-proposed biomechanical features from the finals of long jump competitions in the World Championships. The objectives of the analysis include identifying the most important features contributing to top-performing jumps and exploring the combined effects of these key features. Using quantile regression, we model the relationship between the biomechanical feature set and the target variable (effective distance), with a particular focus on elite-level jumps. To interpret the model, we apply SHapley Additive exPlanations (SHAP) alongside Partial Dependence Plots (PDPs) and Individual Conditional Expectation (ICE) plots. The findings reveal that, beyond the well-documented velocity-related features, specific technical aspects also play a pivotal role. For male athletes, the angle of the knee of the supporting leg before take-off is identified as a key factor for achieving top 10% performance in our dataset, with angles greater than 169°contributing significantly to jump performance. In contrast, for female athletes, the landing pose and approach step technique emerge as the most critical features influencing top 10% performances, alongside velocity. This study establishes a framework for analyzing the impact of various features on athletic performance, with a particular emphasis on top-performing events.
BibTeX
@misc{gan2025feature,
  author = {Qi Gan and Stéphan Clémençon and Mounîm A. El-Yacoubi and Sao Mai Nguyen and Eric Fenaux and Ons Jelassi},
  title = {{Feature Impact Analysis on Top Long-Jump Performances with Quantile Random Forest and Explainable AI Techniques}},
  year = {2025},
  eprint = {2508.09810},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2508.09810},
}
Figure