Football Match Prediction Classifier using data from over 30 years from the top European leagues to predict future match results. The project involves: EDA, data cleaning, feature engineering, feature selection, training and optimising multiple classification models for the best accuracy.

• Used pandas and seaborn to analyse the dataset and perform some initial analysis, and found some trends in the dataset
• Used pandas to clean some corrupted data
• There is a clear relationship between the home team tending to score more goals per game across all the different leagues

• Used pandas to join 3 separate datasets together
• Cleaned the dataset by filling in missing data
• Created a number of new features, ensuring no data leakage:
  • Form
  • Total goals
  • Discipline

• Created a data pipeline which cleans the dataset and uploads it to a SQL database
• Created a feature engineering pipeline to generate new features from the cleaned dataset
• The new features generated can be altered and tuned by hyper-parameters

• Trained a baseline Logistic Regression model to predict the result: Home Win, Draw or Away Win
• Used Random Forest and LightGBM models to improve performance
• Looked at which features can be removed to improve performance (Leave One Feature Out - LOFO)
• Looked at which years of the dataset which can be removed in order to improve performance
• Saved the final model as model.pkl

• Scraped data for matches that have been completed since the original dataset was put together
• Used the data cleaning and feature engineering pipelines to transform the new data
• Used the model to make predictions for these new datapoints
• See below for the results for Premier League 2023-24 (Gameweek 12, w/c 22nd Oct)