kenny613 / football-players-market-value-prediction

A data mining project to predict the market value of football playera

Market value prediction of football players

Introduction

This is a data mining project that aims to mine real footballing data to extract useful knowledge using data mining and machine learning techniques. There are 4 main sections in this report: Problem descriptions, Dataset and preprocessing, Algorithms, Results and analysis. The overall data mining procedures and techniques used are also introduced.

Therefore, the objective of this project is to use real in-match statistics to predict the market value of the football player in € in the sense that we can analyze the goodness as well as the level of the football players.

Data sources

1. FBref.com
2. Transfermarkt.com

Prediction models

• Neural network
• Ridge regression
• ElasticNet

Dimensionality reduction approaches

• Principal Component Analysis (PCA)
• Autoencoder
• Feature Selection

Results

Syntax	MAE mean	MAE min	RMSE mean	RMSE min
PCA+NN (Without log transformation)	7.55715625	7.0544	12.46855625	11.4937
Autoencoder + NN (Without log transformation)	8.480625	7.1313	12.36180625	7.3327
PCA+NN (With log transformation)	7.85724375	6.363	14.41860625	11.1701
Autoencoder + NN (With log transformation)	8.4075875	7.0791	13.9653375	8.7817
Ridge Regression	6.3427	5.8817	11.6661	10.0278
ElasticNet Regression	6.41438	5.9112	11.6596	10.4843