Using machine learning to predict the score of the soccer players of Campeonato Brasileiro de Futebol and decide the best squad with linear programming

About The Project

This is a project to try to build the best team in Cartola FC. To do that, I did first a regression of the score of the soccer players, second I predicted the winner teams and losers of each round using classification with the program that I made in the repository Machine-learning-soccer-classification to prevend the linear model to choose players from teams that are playing against each other which is bad most of the time and for last I used linear programming to decid the best squad maximizing the score predictions multiplied by some weights like, if the team is playing home or outside, if the teams are playing a classic and the position of the team in the Brasileirão table.

Data

I used the data from the repository caRtola: https://github.com/henriquepgomide/caRtola

Built With

Python in Jupyter Notebook

I decided to try different machine learning approaches to compare the results, so i used XGBoost, Extra Trees, Random Forest and MLP.

Features and scores

The meaning of each feature and its weight can be found in the site: https://www.cartolafcbrasil.com.br/scouts

Data for training and testing

train = 70% of data and test = 30%

The weight of playing home

The graph show us that playing home is very important for the Campeonato Brasileiro, because the teams win more often home than outside and that's why i added a weight of playing home in the linear programming part.

Results

Using data from round 14 of 2021, the classification predictions are:

Using 108 cartoletas, the best squad chosen was:

Total price = 107.84

Pontuation prediction = 104.13