Prolego: A Data Science Approach to Predict the Outcome of a Football Match

Abstract

Prolego aims to predict results of Premier League football matches accurately by applying machine learning techniques to historical data. The historical data consists of rows where each row consists of several statistics for both the Home Team and the Away Team. The historical data is generated using web scraping libraries such as Selenium and BeautifulSoup. Based on the scraped data, data cleaning and feature engineering is done to generate several features of a football match like Shots, Shots On Target, Possession, Tackles, Corners, Ratting etc. Finally, the features are represented in a vector format and fed as inputs to different Machine Learning classifier algorithms like Multinomial Logistic Regression, SVM, Gradient Boosting Classifier and DecisionTreeClassifier. After the classification, accuracy is measured by calculating percentage of correct predictions and percentage of correct draw predictions. Error analysis is performed using techniques like Region under Curve to tune hyperparameters and identify the features which are more prominent/useful in accurately predicting the results.