ritviyer / fpl-team-prediction

Fantasy Premier League prediction using Machine Learning

FPL-Team-Prediction

This project uses machine learning to predict the fantasy premier league performance of each player.

• current year/ - Current season's FPL data of each player
  
• data/ - Historical FPL data of each player
  
• prediction/ - Each gameweek's prediction data as well as model predictions
  
• training-data_2016-19/ - Training data from previos seasons
  
• CalculatingFunctions.py - Functions to shift rows, fill Nan values, calculate average of features in last few select gws
  
• FetchFPLData.py - Functions to fetch team and player data using FPL APIs - (Fixture, gw data, player history, etc)
  
• FetchUnderstatData.py - Functions to fetch team and player data from understat - (player xG, xA, team results, offensive/defensive form, etc)
  
• GBMFPLModel.py - Build an XGBoost model for points prediction
  
• LinearRegressionFPLModel.py - Build a Linear Regression model for points prediction
  
• RandomForrestFPLModel.py - Build a Random Forest model for points prediction
  
• GetTeamPoints.py - Fetch gameweek points of each player using FPL APIs
  
• MapUnderstatToFPL.py - Functions to match and Consolidate the player's FPL and understat data
  
• PickTeam.py - Use Linear Programming to pick optimal team
  
• PreparePredictionData.py - Gather data for each player to make a prediction for the following gameweek
  
• PrepareTrainingData.py - Gather historical data for each player for training the models
  
• ReadFPLData.py - Read and process the collected FPL data of each player

For each gameweek, the models are trained using all historical data prior to that week starting from the 2016/17 season. Some of the features used are:

• Player performance in the season so far, ex - goals, assists, clean sheets, bonus points, minutes played, etc
• Understat player data such as xG, xA and more
• Offensive and defensive form of the player's team
• Offensive and defensive form of the opponent team

After points forecast, Linear Programming is used to find an optimal team.

A comparison of the performance of all models will be added at the end of the season.

Acknowledgments

Sincere thanks to:

• https://github.com/vaastav/Fantasy-Premier-League for providing historical FPL data of each player
• https://github.com/amosbastian/understat for providing a Python package for Understat

fpl-team-prediction's Issues

Extension - Team prediction

Hi, the project looks quite interesting.
What would you change to make this work for Serie-A instead? Is all the training data downloaded automatically at some point?