FiFA-World-Cup-Challenge

Can you predict which team will win FIFA World Cup Qatar 2022?

(Source: https://www.fifa.com/fifaplus/en/tournaments/mens/worldcup/qatar2022)

Abstract

FIFA World Cup [1] is the most famous football competition in the world and it will be held from Nov 20, 2022 to Dec 18, 2022 [2]. In this project, we want to predict results of the quarterfinals as well as the exact goals scored by each team using Machine Learning algorithms. Aiming at those goals, we need to leverage massive amounts of data from historical tournaments of each team and ratings of individual players, which is well-suited for machine learning approaches. Specifically, predicting results can be modeled as a classification problem and predicting goals scored by each team can be modeled as a regression problem. To evaluate our method, we will use data from FIFA World Cup 2018 as validation and finally test prediction accuracy on FIFA World Cup 2022 quarterfinals.

Prerequisites

1. Prepare the prerequisites

   pip install -r requirements.txt

2. The directory structure of this project is shown as below:

   .
   ├── README.md
   ├── data
   │   ├── V1
   │   ├── V2
   │   └── raw_data
   ├── data_preparation.py
   ├── models
   ├── models.py
   ├── pic
   ├── requirements.txt
   ├── train.py
   └── utils.py

Data

1. This project has collected and preprocessed the soccer datasets from Kaggle, e.g., FiFA World Cup, FIFA Rankings, Player Score, International Matches . It is recommended to put them under directory data/raw_data.

2. To preprocess the data:

   python data_preparation.py

   Arguments are shown as below:

   Argument	Default	Description
   --data_dir	'./data/V2'	Data directory to save.
   --raw_data_dir	'./data/raw_data'	Raw data directory.
   --cur_year	2018	Test data year.

P.S. In our V1 experiments, we observed the inbalance issues of feature importance. To alleviate this problem, we bring more relevant features in to our V2 dataset and achieved better results.

Usage

1. Train and evaluate ML models on WorldCup dataset. For example,

   python train.py --model lgb

   Some important arguments (listed below, more could be found in train.py):

   Argument	Default	Description
   --model	lgb	ML models for training and evaluation: e.g., KNN
   --data-dir	'data/V2'	Directory path for the data set.
   --model-dir	'models'	Directory for saving trained model files
   --cur-year	2018	Test set year.

Results

1. Results for regression-based model

   Model	Scale	Test MSE on V1	Test R2 on V1	Test Acc on V1	Test MSE on V2	Test R2 on V2	Test Acc on V2
   PR	Standard	2.1654	-1.1983	0.5	1.4530	-1.2368	0.5
   RF	None	2.2198	-1.2657	0.25	1.4877	-1.2650	0.25
   GB	None	2.1726	-1.2065	0.25	1.9732	-0.9924	0.5
   KRR	Standard	2.1813	-1.2546	0.5	1.9313	-1.0260	0.75
   LGB	MinMax	2.0754	-1.1583	0.5	1.3259	-0.1720	0.5

2. Results for classification-based model

   Model	Scale	Test Acc on V1	Test Win Acc on V1	Test Acc on V2	Test Win Acc on V2
   KNN	None	0.125	0.75	0.125	0.75

3. Feature importance analysis

   

Enjoy the World Cup 2022!

References

[1] Nam-Su Kim and Laurence Chalip. Why travel to the fifa world cup? effects of motives, background, interest, and constraints. Tourism management, 25(6):695–707, 2004.

[2] Wikipedia. FIFA World Cup — Wikipedia, the free encyclopedia. http://en.wikipedia. org/w/index.php?title=FIFA%20World%20Cup&oldid=1111687599, 2022. [Online; ac- cessed 30-September-2022].