Soccer Match Predictor - Introduction Rough Draft

Soccer Gif

Team members: Anna, Elshiekh, Alan, Yotam

Introduction

If you follow soccer or any sport for that matter, religiously, you’ll quickly realize that the outcomes are as unpredictable as guessing lottery numbers. Furthermore, accurately predicting the outcome of soccer matches is difficult. Last year, 1.5 billion people watched the world cup, making soccer the most popular sport in the world. Fans around the world try to predict the outcome of soccer matches for fun or even to make money through sports betting. As such, it is important to create a machine-learning model that can accurately predict the outcome of soccer matches.

You can always have a good idea of who might win based on team strength, but even that cannot be trusted at times due to the many other factors that come into play: inconsistency of players, unpredictable events during and surrounding a match, the dynamic nature of team performance and luck. Traditional statistical models lack the capability to incorporate various factors and rely heavily on past performance without considering current conditions. Furthermore, bias and limited capacity to analyze multiple factors in real-time add to the difficulty. Machine Learning has the potential to build relationships across many factors and potentially improve some of these difficulties.

Using an artificial neural network, we will use various factors such as player statistics, team dynamics, and performance in previous matches for our prediction. The integration of diverse data sources and the use of advanced machine learning algorithms will help overcome the limitations of prior work. The dataset used contains information about European soccer teams from 2008 to 2016. The Kaggle dataset contains information from 25,000+ matches and is a very population dataset for soccer. information.

In our results, we expect to see significant improvement in prediction accuracy compared to traditional statistical models and human intuition. We test our results by comparing our model’s predictions against actual match outcomes and against other prediction systems. In our future work, we would like to make regular updates to our model to keep it current with new information and features as well as potential advancements in machine learning techniques.

Related Works

There is a variety of studies that use neural networks to predict the outcomes of soccer matches. One such study is “Predicting Soccer Match Results in the English Premier League” by Ben Ulmer. The study aimed to predict the results of soccer matches in the English Premier League using machine learning models. The authors faced challenges such as the randomness of soccer data and the lack of injury information for players. Features such as whether a team is a home or away, team Elo rankings, and recent results were used for feature selection. The authors also tested various models including a stochastic gradient descent algorithm, Naive Bayes, a hidden Markov model, and a Support Vector Machine (SVM) with an RBF kernel. They managed to achieve about 50% accuracy when predicting whether the match was a win loss or a tie. The authors suggested that more advanced methods of feature selection and data preprocessing may improve the accuracy of the predictions.

Another study that uses more common methods is “A deep learning framework for football [soccer] match predictions” by Ashiqur Rahman. This article, written by Ashiqur Rahman, focused on predicting the outcome of international soccer matches with a focus on the 2018 World Cup. The model was able to accurately predict 63.3% of outcomes correctly, however the author did mention that this number could be increased through more and accurate information on the teams. Some limitations to consider were that the model was able to accurately predict the majority of group stage matches, but failed quite often when it came to quarterfinal, semi-finals, and final matches; this may be because in the later stages teams tend to better matched up with each other, so the model was not be able to label a substantial difference in each team’s possibility to win.

A study that uses more advanced methods is “Predicting Wins, Losses and Attributes’ Sensitivities in the Soccer World Cup 2018 Using Neural Network Analysis” by Amr Hassan et. al. This paper analyzes data from the 2018 World Cup to create a model for predicting match results using supervised learning. The model is based on the “Radial basis function using 75 attributes”. The authors were able to achieve a win rate of 83.3% and a loss rate of 72.7% using key attributes such as Total Team Medium Pass Attempted (MBA) and the Distance Covered Team Average in zone 3. Additionally, the study “A Comparative Study on Neural Network Based Soccer Result Prediction” by Burak Galip Aslan and Mustafa Murat Inceoglu uses the black-box approach to achieve better prediction results. Traditional methods of predicting soccer matches rely on statistical models, however, the black box approach has proven to be better. The authors of the paper compared different models, including one by Cheng et. al. that uses a hybrid neural network, and their black box method was significantly better than the other methods. One thing the paper highlights is the importance of input data. More specifically, they note ‘The available data should be transformed into alternative formats… It may not be necessary to apply any available data in the form of input parameters’ (Aslan and Inceoglu). With this method, the authors of the paper were able to achieve correct home win predictions 70-80% of the time. Tiwari et. al, authors of “Football Match Result Prediction Using Neural Networks and Deep Learning” were able to achieve similar results. This paper uses data from the 2010-11 through 2017-18 seasons of the English Premier League. It employs recurrent neural networks with LSTM cell modifications to predict the outcome of soccer matches. This allowed them to achieve success percentage of 80.75, which is a 10% increase from using ANN.

“Neural Networks Football Result Prediction” by E. Tjioe , F. Syakir , R. H. C. Shum, I. Buo, students at the University of Tartu, used their model to test betting returns. They used data from the English Premier League and the Spanish La Liga. They first created baseline models using other machine learning algorithms, like logistic regression and random forests. They then built a neural network and tuned its hyperparameters to predict the winner of each game. They showed that betting for winner, home or away, and predicted winner with threshold returned greater profit than betting with draw only factors.

Methods

We obtained the dataset from kaggle and need to make sure it looks clean and take care of some preprocessing logistics. Of the 25,000 data points collected from different leagues, we used matches from countries in the top 5 leagues: England, France, Germany, Italy, and Spain.

We also removed data that was formatted in XML and the predicted odds from various betting companies. Our dataset then contains the ids for the country, home and away team, total goals scored, home wins and away wins percentage, and a column of 0s and 1s where 1s represent a home team win.

We are using pyTorch’s Neural Network library, from which we will most likely be using the feed forward architecture. We will be creating different models and comparing the accuracy of using different features.

We are using pyTorch’s Neural Network library, from which we will most likely be using the ANN architecture.

For our model, we are hoping to train the neural network to accurately predict whether Team A or Team B will win, or if the result will be a tie.

The end result we are hoping for is a model that when asked for the outcome of a soccer match would seem like a very well informed sports analyst.

Possible pitfalls we see in our model is low or inaccurate classification.

Discussion

We used the European Soccer Database and filtered out the teams that did not belong to England, France, Italy, Spain, or Germany. We created 2 models, a simple feedforward neural network that only considers the goals scored by both teams and a more complex model that factors in home win percentage, face-to-face history, and player stats. It should be noted that player stats are often not a good measure for predicting the outcome. Roland Shum in Neural Networks Football Result Prediction points out that player morale, a stat that cannot be measured, often plays a role in determining how a player does in a given match. Our comparison will focus on the accuracy of these models to those of other models built. Running the simple model produces an accuracy of ~55%. Compared to our related works our results were significantly worse. We yielded an accuracy of below 60% for our validation dataset while many optimized models in the field can consistently predict with above 70% accuracy. We are intending on using a classification model architecture and building different combinations of features for further experiments.

Works Cited

B. G. Aslan and M. M. Inceoglu, “A Comparative Study on Neural Network Based Soccer Result Prediction,” Seventh International Conference on Intelligent Systems Design and Applications (ISDA 2007), Rio de Janeiro, Brazil, 2007.

E. Tiwari, P. Sardar and S. Jain, “Football Match Result Prediction Using Neural Networks and Deep Learning,” 2020 8th International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO). 2020.

Hassan A, Akl A-R, Hassan I, Sunderland C. Predicting Wins, Losses and Attributes’ Sensitivities in the Soccer World Cup 2018 Using Neural Network Analysis. Sensors. 2020.

Rahman, M.A. A deep learning framework for football match prediction. (2020)

Shum, Roland. “Neural Networks Football Result Prediction” 2020.

Ulmer, Benjamin and M. Pasadas Fernández. “Predicting Soccer Match Results in the English Premier League.” 2014.