Identification of Features for Profit Forecasting of Soccer Matches

For the increase of profit in soccer clubs, it is crucial to understand which are the most relevant features that led to profitable soccer matches. The model aimed to identify the most important explanatory features to predict the profit of soccer matches, using databases from historical soccer matches of Sport Club Corinthians Paulista, as well as public data sets provided by Brazilian government agencies. We applied several classification algorithms to chose the ones that led to the highest ROC AUC (Area Under the Receiver Operating Characteristic Curve). Then we performed feature selection algorithms: LASSO (Least Absolute Shrinkage and Selection Operator) and RFE (Recursive Feature Elimination), to highlight the relevant features from the data set. The top classifiers were Gradient Boosting and Random Forest, which resulted in ROC AUC of 0.860 and 0.842 respectively. The number of features selected by the LASSO and RFE was, respectively, 14 and 17, of which only 10 features were common for both selectors, and, thus chosen as the most important features.