📅 𝗖𝗼𝘂𝗿𝘀𝗲: 𝗗𝗮𝘁𝗮𝗧𝗮𝗹𝗸𝘀 𝗖𝗹𝘂𝗯 - 𝗠𝗮𝗰𝗵𝗶𝗻𝗲 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝗭𝗼𝗼𝗺𝗰𝗮𝗺𝗽 𝟮𝟬𝟮𝟰
For my final project in the Machine Learning Zoomcamp 2024, I applied deep learning to predict match outcomes in the Premier League, aiming to build a more sophisticated model than traditional machine learning methods.
🔍 𝗣𝗿𝗼𝗷𝗲𝗰𝘁 𝗚𝗼𝗮𝗹:
• Predict football outcomes (home win, draw, away win) based on historical match data.
• Explore the performance of deep learning models.
• Gain insights into team dynamics and strategies through the lens of data science.
🧠 𝗧𝗲𝗰𝗵 𝗦𝘁𝗮𝗰𝗸 &amp; 𝗧𝗼𝗼𝗹𝘀:
• Deep Learning: TensorFlow, Keras, LSTMs, CNNs, Keras Tuner for hyperparameter tuning.
• Data Science Libraries: Python, Pandas, NumPy, Scikit-Learn.
• Infrastructure: Docker, AWS Elastic Beanstalk, Streamlit.
• Version Control: Git, Pipenv for managing dependencies.
🔧 𝗞𝗲𝘆 𝗦𝘁𝗲𝗽𝘀 &amp; 𝗧𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝗗𝗲𝘁𝗮𝗶𝗹𝘀:
𝟭. 𝗗𝗮𝘁𝗮 𝗔𝗰𝗾𝘂𝗶𝘀𝗶𝘁𝗶𝗼𝗻 &amp; 𝗣𝗿𝗲𝗽𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴:
• Collected Premier League data (2005-2025) using Football Data API.
• Engineered features such as rolling averages, goal differences, team rankings, and betting odds to provide a comprehensive view of team performances.
𝟮. 𝗠𝗼𝗱𝗲𝗹𝗶𝗻𝗴:
• Dense Neural Networks for baseline comparison.
• Convolutional Neural Networks (CNNs) for spatial feature extraction.
• Long Short-Term Memory Networks (LSTMs) are used to capture temporal dependencies in match outcomes over time.
• Temporal Convolutional Networks (TCNs) to leverage sequential information and improve predictions.
• LSTM with Attention Mechanism: Enhances the LSTM model by focusing on important parts of the sequence
• Utilized Keras Tuner for hyperparameter optimization, improving model performance.
𝟯. 𝗘𝘃𝗮𝗹𝘂𝗮𝘁𝗶𝗼𝗻 &amp; 𝗕𝗮𝗰𝗸-𝗧𝗲𝘀𝘁𝗶𝗻𝗴:
• Validated models using cross-validation and back-testing on historical match data.
• Compared the model’s predictions against market betting odds and Brier score metrics to assess its predictive accuracy.
• Achieved valuable insights, particularly in teams where the deep learning model outperformed market expectations.
𝟰. 𝗗𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁:
• Deployed the final model to AWS Elastic Beanstalk for scalable, cloud-based inference.
• Containerized the application using Docker for easy deployment and reproducibility.
• Developed a Streamlit app to provide interactive predictions for match outcomes.
🔗 𝗙𝗶𝗻𝗮𝗹 𝗗𝗲𝗹𝗶𝘃𝗲𝗿𝗮𝗯𝗹𝗲𝘀 &amp; 𝗚𝗶𝘁𝗛𝘂𝗯:
The project is available on GitHub: <a href="https://lnkd.in/dPstKqdA" target="_blank" rel="noopener nofollow">https://lnkd.in/dPstKqdA</a>
#MachineLearning #DeepLearning #FootballPrediction #DataScience #ML #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

🚀 𝗙𝗶𝗻𝗮𝗹 𝗣𝗿𝗼𝗷𝗲𝗰𝘁: Premier League Football Prediction using Deep Learning