Background

Football is one of the most popular sports worldwide, winning a football match has become one of the most crucial aspect of football clubs.In this project, we will be analyzing the matches across different competitions (leagues & cups) through the API.

In recent decades, football clubs have been publicly traded and has become an investment target by many. Predicting match outcomes have become very important to generate a positive return on investment. The data provides the pre-match odds of winning a game (whether the home team or away team will win, or draw). It also provides team names, halfway scores, and fulltime scores. We are also including FIFA’s team ratings in our dataset. Using this data, we are trying to predict the outcome of the matches.

All about Data!!!

• Primary Source of our data: Football API https://www.football-data.org/documentation/quickstart

• Secondary Sources:

  • Additional match data from http://football-data.co.uk/data.php
  • Performance Rating Data from https://www.fifaindex.com/teams/

Original Plan

• Our goal is to predict the outcome of matches using the pre-match odds of winning a game

Analytics Plan:

• Visualize the data in R
• Examine the wins for home and away matches
• Explore different variables and choose the best factors for model training
• Predict wins for using different machine learning models such as Logistic regression, Support Vector Machine, and KNN

Current Objective

• To predict winner after full-time using the score in Half-time, pre-match odds and performance rating

Analytics plan:

• Exploratory data analysis of matches across different competitions
• Consider not only pre-match odds but half-time scores as well as team Performance Ratings
• Use H2o’s AutoML for automating the machine learning workflow, which includes automatic training and tuning of many models within a user-specified time-limit.

Summary of Peer Comments

Many teams suggested us to:

• Include weather data; the data was not specifically available for matches in the API or other external resources
• Use player level analysis, such as player being injury prone, ranks, etc.; we did not get Player attributes but we included Team Performance in our dataset
• Perform sentiment analysis on the live tweets for the matches; our project focuses on predicting the wins, however this could altogether be a different project

The feedbacks provided by some of the teams allowed us to think beyond our limits. Suggestions Incorporated:

• Considering Team/Player’s performance ratings
• Predicting for 11 Leagues
• Using H2O’s AutoML over traditional ML models

Data Cleaning & Transformation

• This is the raw data about the matches that we obtained from our API

Data Cleaning & Transformation

• This is the data post unnesting the raw data and selecting relevant columns

• Since we had limited data available from the API, we included external data from (http://football-data.co.uk/data.php) to enhance our dataset

• We dropped the NA values from the combined dataset and came up with 20K+ rows

Data Exploration

Data Exploration

Adding Team Performance Ratings

• For our analysis, We are also using performance ratings of the teams from (https://www.fifaindex.com/teams/) in order to enrich data.

• Finally we have a dataset of matches containing pre-match odds, half time and full time goals as well as performance ratings of the team

H2O’s AutoML

• We have used H2o’s AutoML for automating the machine learning workflow, which includes automatic training and tuning of many models within a user-specified time-limit.

Results & Conclusion