Soccer Games Results Prediction ECE 539 Introduction to
- Slides: 10
Soccer Games Results Prediction ECE 539 – Introduction to Artificial Neural Networks and Fuzzy Systems Henrique Parreiras Couto
Background � The first division of Brazilian soccer league includes 20 teams � Every team plays against all others twice � Total of 380 games per year � The championship format was different before 2003
Project Goal � Predict the score of any match of the first division of the Brazilian national championship using a Multi-layer perceptron.
Data Extraction � Public study about the market value of Brazilian teams Source: http: //www. pluriconsultoria. com. br/relatorio. p hp? segmento=sport&id=263
Data Extraction � Publically available game results from 2003 through 2012 � Python program was used to extract and format the data into. txt files according to each team (with Alberto Tavares) � http: //www. bolanaarea. com/gal_brasileirao. htm
Feature Vectors � MATLAB program used to assembly the data � Home Team ◦ ◦ # of matches played since 2003 Home goals for Home goals against Market value � Away ◦ ◦ Team # of matches played since 2003 Away goals for Away goals against Market value
Feature Vectors - Labels � [1 � [0 0 1 0 0 0 0 0 1 0 0] 0] 1] – – – Large loss Small loss Tie Small victory Large victory
Feature Vectors � Training and Testing files ◦ 380 feature vectors each Games Played Home goals for Home Market goals Value against Games Played Away goals for Away goals against Market Value Labels
Score prediction � Classifier result gives the difference between the number of goals of each team � Final score prediction based on the classifier result and average number of goals scored by each team since 2003.
Results � Average classification rate of the MLP : ~40% � Improvements needed
