I’m thrilled to share the progress on my midterm project—a machine learning model that predicts Premier League match outcomes. It’s been an exciting journey diving into data from nearly two decades of English football, building a model that tackles one of the biggest challenges in sports analytics: predicting home wins, draws, or away wins.
<h3>Project Highlights:</h3>
<ol>
<li>
Data Collection &amp; Preparation:
<ul>
<li>We pulled and cleaned Premier League data from 2005 to 2024, engineered features such as goal differences, cumulative points, rolling averages, and normalized betting odds—laying a solid foundation for accurate predictions.</li>
</ul>
</li>
<li>
Exploratory Data Analysis (EDA):
<ul>
<li>Key metrics were analyzed, correlations explored, and data quality checked to ensure meaningful insights for modeling. These initial steps help validate and refine our dataset for improved accuracy.</li>
</ul>
</li>
<li>
Model Training:
<ul>
<li>Using Random Forest and XGBoost models, we optimized feature selection and hyperparameters to manage overfitting, resulting in a model with solid generalization. Evaluation metrics include learning curves and Brier scores to ensure model robustness.</li>
</ul>
</li>
<li>
Back-Testing:
<ul>
<li>The model’s accuracy is compared to betting market probabilities, offering unique insights into its competitive edge over traditional forecasts. It’s a great step towards understanding the model’s performance in real-world scenarios.</li>
</ul>
</li>
<li>
Deployment Options:
<ul>
<li>Comprehensive deployment instructions make it easy to set up locally, Dockerize, or deploy via AWS Elastic Beanstalk. There’s even a Streamlit app for interactive predictions, adding a layer of accessibility and engagement for end users.</li>
</ul>
</li>
</ol>
<h3>Getting Started</h3>
The project setup is streamlined: Python 3.8+, Docker, and Pipenv. With dependencies installed, you can easily clone the repository and dive into predictions or even deploy the model for hands-on testing.
Check out the full project here: <a href="https://github.com/RuiFSP/mlzoomcamp2024-midterm-project" target="_blank" rel="noopener nofollow">GitHub Repository</a>
A huge thank you to my fellow learners at <a href="https://www.linkedin.com/company/datatalks-club/" target="_blank" rel="noopener nofollow">DataTalksClub</a> for the collaborative discussions that keep this learning experience enriching and dynamic. And special thanks to <a href="https://www.linkedin.com/in/agrigorev/" target="_blank" rel="noopener nofollow">Alexey Grigorev</a> for creating such a strong, supportive community!
Excited to keep learning and evolving in the field of sports analytics! ⚽📊
#MachineLearning #DataScience #SportsAnalytics #PremierLeague #DataTalksClub

It's a community where individuals openly share their learning journeys, insights, and challenges to help others grow. It’s a supportive space focused on transparency, mutual encouragement, and collective growth

LearningInPublic

A machine learning model has been developed to predict Premier League match outcomes based on data from 2005 to 2024. Key steps include data collection and preparation, exploratory data analysis, model training using Random Forest and XGBoost, and back-testing against betting market probabilities. The project supports deployment via Docker or AWS Elastic Beanstalk, and includes a Streamlit app for interactive predictions.

🚀Week 7 Milestone - Premier League Match Prediction Project Completed! 🎉