Kangwoo Choi, Won Barng, Vincent Widjaya, William Kwok

Sports is an integral part of nearly all cultures. One of the largest sports is football (Soccer. We will refer to it as football). By studying different factors in the data, we can make predictions that many people would be interested in. Team coaches would want to know with data what are the most important factors that lead to a game win. Investors would want to know how likely the team they are sponsoring will do well. Gamblers would like to know gambling trends and what they should bet on. The general public would want to know if their favorite team will win or if there may be illegal match fixing going on with the betting that we cannot see unless we look at the data. As individuals who are passionate about football, we are interested in exploring and predicting outcomes and trying to find hidden patterns.

For years individuals have tried to use statistics to figure out what makes teams win, or try to find out if their favorite teams are the best. Individuals have tried to use statistics to figure out what makes teams win, or to try to find out if their favorite teams are the best. In fact, "the research for predicting the results of football matches outcome started as early as 1977 by [Stefani R]" (Razali et al).

A study done in 1997 by AJ Lee previously explored the Poisson Regression in an attempt to model team scores to determine if Manchester United was the best team in the Premier League. They predict with some good correlation, but they say their "approach to modeling the scores is a little simplistic. [They] have taken no account of the fact that teams differ from game to game due to injuries, trades, and suspensions. In addition, [they] are assuming that [their] model leads to reasonable probabilities for winning/losing/drawing games".

We can improve on the accuracy by including in our model some of the methods AJ Lee suggested to improve on. It will be difficult to truly predict, as complexity of the game "makes the game result hard to be predicted" (Junyuan Gao) , who also attempted to use a Poisson Regression for her predictions.

We want to test if there is a correlation between:

• betting odd and actual result.
• between betting odd and game stats.
• between game stats and actual result.

We will be using data from all England Premier League seasons from the 2014/2015 season to the current 2018/2019 season from Football Data. This dataset contains data pertaining to the game itself as well as a lot of bets associated with the game.

We would use the Poisson Distribution for the number of goals by each team. Also, we will try the SGD Regressor and Lasso regression algorithms as suggested by the sci-kit learn algorithm cheat sheet (https://scikit-learn.org/stable/tutorial/machine_learning_map/index.html). In order to select important features in some models, we will use our domain knowledge of football in addition to forward and backward selection.

Our study can be of interest for a number of target audiences. These include soccer fans and followers, soccer betting enthusiasts, as well as investors of soccer teams. However, we will focus on soccer bettors as they would be most interested in the details of our study to maximize their chance of winning. Unlike the other two, bettors’ purpose that is to win bets relate most to ours, that is to find correlations between betting odds, wins, and player/team statistics.

Some questions our audience could answer from our resource could be:

• What common trends do teams that end up on top have earlier in the season?
• What specific team stats should I focus on watching to maximize my betting odds and rewards?
• Which bookmaker(s) should I bet with that would maximize my rewards?

During the process of data collection/data management, we anticipate facing a few challenges. The first challenge we anticipate to face is not getting the individual player's data. We thought player’s data is also an important factor to predict the winner / final points of the league, our datasets only have each teams’ data. The second challenge we anticipate to face is having trouble with finding dataset related to variables that the dataset doesn’t contain. For instance, we could not get the data about the stadium conditions; the length of lawn or weather condition of the game.

In order to complete the project, we need to learn additional technical skills that we did not learn in class. Since we are going to use Poisson distribution to predict the points and rankings, we need to learn about Poisson distribution packages. Also, if we choose to use a neural network as our machine learning technique, we need to research about the algorithm and learn how to apply in python. Lastly, we want to learn and use data visualizations that we did not use in the lecture.

To analyze the data, we will first think about what features affect the results of the game. We think that betting odds and game stats, especially shoot on target, goals, and red cards, affect the result of the game. Using current stats of the team, we think we can predict the results of remaining games, and predict the ranking. After predicting the variables, we will use forward/backward selection to find the set of variables that improve the linear regression the most. Then, we will use linear regression to find out the relationship between the game result and the features we chose. Lastly, we will use Poisson distribution to predict the result of the remaining games. A Poisson distribution shows the likely number of times that an event will occur within a predetermined period of time. (Tutorials point) Since each team left with 10 games remaining, we can predict the result of remaining games and predict final ranking or points of the teams. After predicting the results of the games, we will convert our paper to HTML and post on the GH-pages on Github.

There are a few major challenges that we anticipate to face. The first challenge would be having a consistent dataset. At the end of the season, last 3 teams are regulated to the second division, and 3 teams from second division promote to the premier league. Since we are looking at the data of the past 5 years, the teams that are regulated or promoted in the past 5 years will have a fewer number of the dataset. Also, there are many unpredicted or hidden variables, such as players injuries, stadium condition, change of managers and weather, can highly affect the result of the game.

AJ Lee: https://amstat.tandfonline.com/doi/pdf/10.1080/09332480.1997.10554791

Junyuan Gao: https://www.stat.berkeley.edu/~aldous/Research/Ugrad/Junyuan_Gao.pdf

Razali et al: https://iopscience.iop.org/article/10.1088/1757-899X/226/1/012099/pdf