search

jamsavio / jobs

MQL4 (MetaQuotes Language 4) are integrated programming languages designed for developing trading robots, technical market indicators, scripts and function libraries within the MetaTrader software.
0 stars 0 forks source link

Predicting Football With Data from FIFA Football Game #8

Closed jamsavio closed 4 years ago

jamsavio commented 5 years ago

Dataset inspired:

Shin and Gasparyan [8] conducted a research on predicting football results by mixing real data and data collected from football video game, FIFA 2015. They integrated real data with the players’ attribute data they collected from FIFA 2015 such as heading, passing, shooting, and strength, to predict matches results and it shows better prediction results compared to only real data. They also stated that the method of using data from video games can save a lot of time and energy as it can be very expensive for some data to be calculated or collected from the real world.

Model used:

In this paper a logistic regression model is built to predict matches results of Barclays' Premier League season 2015/2016 for home win or away win and to determine what are the significant variable to win matches. Logistic regression is a classification method which can be used to predict sports results and it can gives additional knowledge through regression coefficients. The variables used are “Home Offense”, “Home Defense”, “Away Offense”, and “Away Defense”.

Benchmark:

Snyder [9] conducted a research to predict Barclays’ Premier League season 2011/2012 matches by using all matches in season 2010/2011. He used a lot of variables, stadium capacity, distance traveled by a team before match, and statistics of a team on previous season, including ranking, amount of wins, draws, losses, scored and conceded goals, goals difference in each match, points, money spent on players’ wages, money spent on 2011 summer transfer market, and number of games a manager of the team has played in the league. He built his model with logistic regression and its prediction accuracy was 51.06%.

Conclusion:

Our work has shown quite significant improvement compared to [9], with our predictive accuracy hitting 69.5% compared to his accuracy on 51%, and we succeed in building a model with less variables but stronger prediction accuracy.

paper: https://ieeexplore.ieee.org/document/7803111