Football-Player-Performance-Analysis

Football is a team sport where each and every decision taken by a single player would affect the performance of the whole team. A minute mistake made by a team member would adversely affect the results of the game. Therefore, player performance analysis should be considered as an important part of the coaching system. Several ranking systems have already been implemented for analysing the performance of the football players. Most of these methods require manual input by the coaches and the team staffs. The problem with this approach is that it leads to a more biased data since there might be favouritism among people. As a result of this, some players might get ranked higher than they deserve while on the other hand, several others would not get featured despite portraying their terrific skills and outstanding gameplay. This eventually results in a terrible analysis of the football player performance. Therefore, there is a need for a system to properly analyse the performance of the football players that can identify their skills accordingly. It should be able to accurately predict the ratings of the different football players based on their skillset and gameplay.
The main goal for this project was to analyse the performance of the football players based on several attributes that are used to measure the footballing abilities of the players. The players were then assigned a rating depending on his/her performance that was calculated using a machine learning model. The various functionalities implemented in the project could be categorised into three main sections. For the first two sections, Lionel Messi was chosen as the football player of preference. The first section involved analysing the changes in the performance of a player with respect to his age. The second one involved finding players which have a playstyle similar to the chosen player. In the third section, BorutaShap feature selection was implemented for finding the important features that contribute towards the prediction. After the feature selection step, six machine learning models were selected for predicting the rating of the football players. For this project, the models of choice were Linear Regression, Decision Tree, Random Forest, K Nearest Neighbour, Support Vector Machine and XGBoost. The performances of these models on predicting the rating were compared using cross validation. Upon acquiring the better model among them, which was XGBoost, the hyperparameters of the model were fine-tuned using Grid- SearchCV to obtain the best model for prediction. The model was then evaluated on the test dataset and a confidence interval was also procured with a confidence of ninety five percent.