Analysis of Table Tennis Service Returns

Project Overview

This project aims to analyze table tennis match data, focusing on service returns. The goal is to group these returns into clusters using machine learning methods, then interpret these groupings to better understand player strategies and behaviors depending on the type of serve received.

The analysis is based on a database provided by LIRIS, containing in particular:

• The coordinates of bounces,
• The hitting positions,
• The clusters of bounces,
• And many other details from numerous professional matches.

---

Main Features

1. Data Display and Exploration

• Notebook 1_display_data.ipynb
  Visual exploration of table tennis match data:
  • Interactive and static visualization of bounces and shots on the table,
  • Display of isolated point videos,
  • Exploration of bounce clusters,
  • Use of Plotly figures for analyzing positions and trajectories.

2. Calculation and Display of Domination and Pressure Indicators

• Notebook 2_domination_pression.ipynb
  Calculation and visualization of two advanced metrics:
  • Domination: a dynamic indicator taking into account the score, set difference, winner of the previous point, and match evolution,
  • Pressure: an indicator measuring the pressure felt at each point (score gap, key moments, end of set, decisive set),
  • Graphical display of the evolution of these indicators throughout the match.

3. Analysis of Return Shot Types Distribution

• Notebook 3_return_type_analysis.ipynb
  Statistical analysis of service return styles:
  • Distribution of different return types (topspin, flip, push shot, block) by player,
  • Visual representation using stacked horizontal bars,
  • Ranking of players based on their offensive return tendency (topspin rate),
  • Comparative analysis of playing styles and tactical preferences in service returns.

4. Cluster Analysis using Elbow Method

• Notebook 4_elbow_method.ipynb
  Advanced clustering analysis of service returns:
  • Implementation of the elbow method to determine optimal number of clusters,
  • Outlier detection and removal using z-score standardization,
  • Visualization of clustering results,
  • Statistical analysis of cluster characteristics (centers, standard deviations),

---

Dash Application

A Dash application has been developed to facilitate data exploration.
Bounces are displayed on a Plotly figure:

• Clicking on a point displays the video of the corresponding rally,
• Several filters allow you to select a player, handedness, or shot number.

---

Advanced Analysis: Domination and Pressure

Domination Indicator

An advanced indicator has been developed to quantify domination during a match, taking into account:

• The score,
• The set difference,
• The winner of the previous point,
• The dynamic evolution of the match.

Pressure Indicator

Another indicator measures the pressure felt by a player at each point, combining:

• The score gap,
• Key moments (set point, match point),
• End of set,
• Decisive set.

---

Installation

1. Clone this GitHub repository:

   git clone https://github.com/centralelyon/table-tennis-returns
   cd table-tennis-returns

2. Install LaTeX dependencies (required for figures):

   • Windows:

     • Install MiKTeX

   • Linux (Ubuntu/Debian):

     sudo apt-get install texlive texlive-latex-extra dvipng

   • macOS:

     brew install texlive dvipng

3. Install Python dependencies:

   pip install -r requirements.txt

---

Authors

• Riad ATTOU : attou.rd@gmail.com
• Marin MATHE : mathe.marin22@gmail.com
• Aymeric ERADES : aymeric.erades@ec-lyon.fr
• Romain VUILLEMOT : romain.vuillemot@gmail.com