Men’s Tennis Performance Analysis using Machine Learning

Project Overview

This project aims to predict the match outcomes for Carlos Alcaraz in the year 2024 using historical tennis match data from 2020 to 2023. Advanced machine learning techniques are employed to forecast win probabilities for upcoming matches, providing valuable insights for coaches, analysts, and fans.

Setup Instructions

Virtual Environment Setup

To ensure that your project dependencies are managed properly and to avoid conflicts with other projects, it is recommended to use a virtual environment. Below are the steps to create, activate, and install the required dependencies.

1. Clone the Repository

First, clone the repository to your local machine:

git clone https://github.com/jg2012/Capstone_Project
cd Capstone_Project


#### 2. Create a Virtual Environment
Create a virtual environment in the project directory. You can use venv for this purpose.

For Windows:
```bash
python -m venv venv

For macOS and Linux:

python3 -m venv venv

3. Activate the Virtual Environment

To activate the virtual environment, run the following command: For Windows:

venv\Scripts\activate

For macOS and Linux:

source venv/bin/activate

4. Install Dependencies

Once the virtual environment is activated, install the project dependencies using pip:

pip install -r requirements.txt

Deactivate the Virtual Environment

When you are done working in the virtual environment, you can deactivate it by simply running:

deactivate

Additional Information

To add new packages to the project, use pip install package-name and then update the requirements.txt file with the new dependencies:

pip freeze > requirements.txt

Data Sources

Data for this project was gathered from multiple sources:

• Sport Radar API: Provides up-to-date player data, including current ATP and WTA rankings, player-specific stats, and match summaries.
• Kaggle Dataset: Historical match data from 2000 to 2023 was obtained from Kaggle.

Key Computational Resources

Several computational resources and tutorials were utilized in this project:

• Python libraries: pandas, numpy, scikit-learn, matplotlib, seaborn, and others.
• Jupyter Notebook for data analysis and visualization.
• GitHub for version control and collaboration.
• Overleaf for LaTeX report generation.

Report and Documentation

The full project report, including detailed methodology, analysis, and results, can be accessed via Overleaf: Project Report on Overleaf.

Repository Contents

• data/: Contains the datasets used in this project.
  • tennis_rankings.csv
  • tennis_rankings_modified.csv
  • tennis_competitor_profile.csv
  • tennis_competitor_profile_modified.csv
  • tennis_summaries.csv
  • atp_tennis.csv
  • atp_tennis_modified.csv
  • Alcaraz_C_matches_2020_2023_best_of_3.csv
• notebooks/: Jupyter Notebooks used for data analysis and model training.
  • ATP_notebook.ipynb
• scripts/: Python scripts used for data processing and model training.
  • data_collection.py
  • data_cleaning.py
  • model_training.py
  • visualizations.py
• img/: Contains all the generated visualizations used in the report.
  • Avg_Set_Score_Per_Year.png
  • 
  • Violin_Plot_of_each_Set.png
  • 
  • Win_and_Loss_Dist.png
  • 
  • win_probabilities.png
  • 
  • feature_importance.png
  • 
  • win_probability_distribution.png
  • 
• requirements.txt: List of dependencies required to run the project.
• README.md: This file.

Class Distribution and Model Performance

Class Distribution Before SMOTE

Outcome	Count
407573	50
352776	2
106755	2
23581	2
47603	2
89632	2

Shape of Data

Dataset	Shape
X_preprocessed	(60, 9)
X_resampled	(300, 9)
X_train	(210, 9)
X_test	(90, 9)
new_X before preprocessing	(48, 2)
new_X_scaled	(48, 9)

Class Distribution After SMOTE

Outcome	Count
407573	50
352776	50
106755	50
23581	50
47603	50
89632	50

Voting Ensemble Accuracy

Metric	Value
Voting Ensemble Accuracy	0.9555555555555556

Confusion Matrix

	23581.0	47603.0	89632.0	106755.0	352776.0	407573.0
23581.0	17	0	0	0	0	0
47603.0	0	14	0	0	0	0
89632.0	0	0	16	0	0	0
106755.0	0	0	0	16	0	0
352776.0	0	0	0	0	11	0
407573.0	0	0	1	2	1	12

Classification Report

Outcome	Precision	Recall	F1-score	Support
23581.0	1.00	1.00	1.00	17
47603.0	1.00	1.00	1.00	14
89632.0	0.94	1.00	0.97	16
106755.0	0.89	1.00	0.94	16
352776.0	0.92	1.00	0.96	11
407573.0	1.00	0.75	0.86	16
Accuracy			0.96	90
Macro avg	0.96	0.96	0.95	90
Weighted avg	0.96	0.96	0.95	90

Conclsion

In this report, we successfully utilized historical data from 2020 to 2023 to predict Carlos Alcaraz's match outcomes for 2024. The ensemble model, combining logistic regression and random forest classifiers, demonstrated robust performance with an accuracy of 95.56 percent. The visualizations provided clear insights into the predicted win probabilities for upcoming matches. These findings offer valuable information for coaches, analysts, and fans, enhancing strategic decision-making for future matches.

Future Work

Future research should integrate additional performance metrics, such as player fatigue, weather conditions, and real-time match statistics, to improve the model's accuracy and robustness. Exploring more advanced models, like deep learning and reinforcement learning, could provide deeper insights and better predictions. Additionally, implementing real-time data feeds and continuous model updating will ensure the model remains relevant and accurate over time. Finally, expanding the analysis to include more players and matches will generalize the model's applicability across the sport.

Result Interpretations

The ensemble model achieved an accuracy of 95.56 percent, indicating a high level of performance. The model shows high precision and recall for predicting wins, with detailed classification reports provided for further insight. The confusion matrix suggests that the model is effective at predicting match outcomes, which is valuable for strategic decision-making.