allen-zengr/football-prediction-web-app

Web app that helps to visualize football predictions. Predictions done with Gradient Boosted Decision Trees.

Football predictor web app

Link: https://football-prediction-web-app.azurewebsites.net/

1. What is this?

This is a web app that helps to visualize football predictions and how each input variable affects the output. Below is an example of a match report. Each bar represents the marginal contribution of the corresponding variable on the output.

2. How was it done?

2.1 Modelling

We use Gradient Boosted Regression Trees to predict the goal difference of a game, Home team goals [minus] Away team goals. Boosted trees are a great choice for this as it performs well with tabular data with non-linear interactions with the output. Below chart explains how the residuals of each tree are fed into each next tree, thus performing gradient boosting:

Source: https://www.geeksforgeeks.org/ml-gradient-boosting/

The algorithm also pairs well with the SHAP package which helps to explain each prediction by assigning marginal impact on the output to each input variable:

Source: https://github.com/slundberg/shap

Below we can see the model performance in terms of accuracy and ROI:

All data taken from: https://www.football-data.co.uk/

2.2 Deployment

• Model training is done offline, inference is done daily with a scheduled data refresh
• Backend is done in Flask
• Frontend is done with vanilla JS, HTML, CSS
• Website hosted on Azure App Service, using Azure Blob Storage for storing data

3. Why was it done?

This web app is primarily for demonstration of data preprocessing, data modeling and some basic web development. The secondary purpose is to use it myself :)

4. Feature glossary

Below is an explanation for each feature in the model:

• Home/Draw/Away odds: European style odds for each outcome. If odds are 1.82 -> you pay 1.00 unit and receive 1.82 in case of a win.
• Goals + (H/A): Cumulative average goals scored
• Goals - (H/A): Cumulative average goals concered
• Conversion (H/A): Shot-to-goal conversion, Goals / Shots on target
• Accuracy (H/A): Shots on target / Shots
• Pts average (H/A): Cumulative average points obtained
• Pts in last 3 (H/A): Average points in the last 3 games
• Table position (H/A): Position in the table, treated as a continuous variable
• Variance (H/A): Deviation of actual result from odds for each team. High variance means team has produced more surprising results, whether positive or negative