Fantasy Premier League Predictor (DSI-SRP)
This DSI-SRP fellowship funded Mubarak Ganiyu to work in the laboratory of Professor David Owens in the Division of General Engineering during the summer of 2020. Mubarak graduated in May 2021 with a major in Mechanical Engineering and a minor in Quantitative Methods.
The project funded by this fellowship aimed to build a web application that is capable of predicting players’ performance so teams can deploy it in signing players inside or outside the premier league. Ganiyu feels that the Premier League is the most entertaining soccer league globally, attracting players from all over the world. Players from different leagues are signed or recruited based on their performance. Ganiyu has built a web application to predict players’ performance in the hopes that teams can deploy it in signing players inside or outside the premier league. The metric used to gauge performance is called FPL points. FPL stands for Fantasy Premier League, an online platform where players are given points based on performance after gameplay. The goal of this project is to be able to calculate the FPL points per game for players in the premier league based on certain stats.
Data was collected from multiple sources, transformed, cleaned, wrangled and combined to build new datasets that were going to be used for data visualization and modeling. The final dataset proved to be very useful, and contained a cumulative weekly data that summed up players’ performance variables from August 2016 till May 2019.
After using different sources to collect data sets, newly formed datasets on 2016/17, 2017/18 and 2018/19 seasons were created. Then, bar charts, scatter plots, box plots and interactive dashboards were built to show how different variables interrelate with each other as well as how the players’ FPL points differed based on positions. Using a dataset that added up players’ stats over the course of three seasons for modeling was optimal. This made it possible for the model to study what happened after 200 minutes of gameplay as well as what happened after 4000 minutes of gameplay and make predictions. Moreover, it was utilizing a huge dataset of over 45,000 records in which each player had about 38 to 114 recorded rows of gameplay in the dataset. This made it easy for the model to learn how different players would be expected to perform based on position and use this knowledge combined with the other systematically chosen stats to predict FPL points.
Ganiyu built a web application that let individuals input players’ names as well as their stats and use the predictive model to output a prediction of the FPL points per game. The values of variables from this dataset can be tested with the web app. To view his website, please visit https://fpl-predictor.herokuapp.com/
In addition to receiving support through a DSI-SRP fellowship, this project was supported and facilitated by the DSI Data Science Team through their regular summer workshops and demo sessions.