Scouting Players with FIFA19
Applying Data Mining to Scouting
Data Driven Approach to Scouting
In the era of eight-figure salaries and nine figure signing fees, player recruitment is a
high-stakes game. In the past, soccer scouts have relied on rudimentary data and intuition to
evaluate the performance and value of soccer players. With the recent rise in data analytics
that can capture many aspects of a player’s performance, statistics and data science are
beginning to play a more prominent role in identifying rising stars and overvalued /
undervalued players.
For this project, we are positioning ourselves as a scouting agency that uses analytics to,
among other things, enhance the discovery of talents and help soccer clubs better understand
the dynamics (features) that come into play when determining the value, overall and future
potential of a player. Our agency will be focusing on solving these fundamental scouting
problems:
1. Finding undervalued players for a given club to acquire,
2. Analyzing a team’s current roster for over-payed and/or underperforming
players that could be traded or sold,
3. Developing a database of similar players for clubs looking for a specific player
type,
4. Build a predictive model to evaluate the future potential of young players.
We will be utilizing the FIFA 19 Player dataset available on Kaggle and apply various Data
Mining techniques to achieve our objectives.
Project Objectives
• Cluster players based various features to identify different player types for our similarity
database.
• Identify under-valued and over-valued players based on ability measures relative to
their value, salary, and/or release clause.
• Building predictive models for future value and potential of players.
Dataset
• Source: Kaggle
• Description: Detailed attributes for every player registered in the latest edition of FIFA
2019 database.
• Size: 9.1MB (18.2k observations x 89 features)
• Features:
1
• ID • Value • Joined
• Name • Wage • Loaned From
• Age • Special • Contract Valid Until
• Photo • Preferred Foot • Height
• Nationality • International Reputation • Weight
• Overall • Weak Foot • Ability by positions (26 features)
• Potential • Skill Moves • Ability by skills (34 features)
• Club • Work Rate • Release Clause
• Position • Jersey Number
Team & Roles
• Markus Wehr: Finding undervalued players.
• Nazih Kalo: Analyzing current roster of players.
• Stephen Stark: Developing similarity database.
• Tam Nguyen: Predictive model for future potential/value.
• Woo Jong Choi: Predictive model for future potential/value.
Data Mining Steps:
• Missing value, data type
Data pre-
• Features distribution
processing
• Feature engineering
1. Pre-processing and EDA
2. Clustering
Analysis
3. Build predictive models
Stages
4. Analyze performance & make final predictions
5. Visualize Output
• PCA
• t-SNE
• K-means
• DBSCAN
• SVD
• Regression: linear/ logit
Potential
• Hierarchical Clustering
Methods
• Latent Class Clustering
• Discriminant Analysis
• Regression Trees
• Random forest
• Decision trees
• Association rules
1. Microsoft Teams
Tools 2. Python
− Jupyter Notebook, Google Collab
2
− Pandas, Numpy, Matplotlib, Seaborn, Scikit-learn, Scipy
3. Tableau
