Predicting the football match winner using LSTM model of Recurrent Neural Networks

This repository contains the code for a conference paper "Predicting the football match winner using LSTM model of Recurrent Neural Networks" that we wrote. This paper gives an introduction to the advantages of using an LSTM (Long Short-Term Memory) Cell in a Recurrent Neural Network and uses it to predict the outcome of a football match.

Dataset

The dataset used here has been obtained from football-data.co.uk. Datasets of the English Premier league have been taken from seasons 2010-11 to 2016-17.

Data Preprocessing

The files dataCleaning.ipynb and fdManipulate.ipynb take the raw data from the website and add attributes regarding the win streaks and the loss streaks for every team. Also, eplStandings.csv contains the final ranks of all the teams in the English Premier League from 2010-11 to 2016-17.

Prediction

The file LSTM.ipynb [Depricated] constructs a RNN using the LSTM cell (tensorflow 1.14) and predicts the outcome of the test dataset.

The file LSTM_New.ipynb constructs a RNN using the LSTM cell (tensorflow keras API) and predicts the outcome of the test dataset.

Result

This model proved to be better than the other models previously used to predict the winner of a football match. Detailed analysis is given in the paper (under review). The accuracy percentages in the paper are incorrect. Please execute LSTM_New.ipynb python notebook to get the correct accuracy.

Libraries Required

1. tensorflow
2. pandas
3. numpy
4. datetime
5. itertools
6. scikit-learn