FIFA 2021 – Comprehensive EDA & Data Cleaning Project

This repository contains a fully educational Jupyter Notebook dedicated to exploring, analyzing, cleaning, transforming, and preparing the FIFA 2021 players dataset for machine-learning applications.
The project demonstrates advanced techniques for data preprocessing, feature engineering, encoding categorical features, handling outliers, and selecting the most important features using several modern approaches.

⚠️ Note: This project is created for learning and training purposes.
It is not intended as a production-ready pipeline but rather as a comprehensive guide to real-world data preparation.

---

📌 Project Overview

The notebook walks through a complete data-science workflow, covering:

• In-depth Exploratory Data Analysis (EDA)
• Cleaning and preprocessing raw messy data
• Handling missing, duplicated, inconsistent, and extreme values
• Converting financial, physical, and textual data into structured numeric features
• Encoding categorical variables using multiple strategies
• Applying transformations (Log, Box-Cox, Yeo-Johnson) to fix skewed distributions
• Scaling and normalization techniques
• Detecting outliers using statistical and multi-dimensional methods (Z-Score, IQR, LOF)
• Selecting important features using Filter Methods, Mutual Information, Correlation, RFE, and RFECV
• Training and evaluating an XGBoost Regression model to predict player overall ratings

---

📂 Dataset

The project uses the FIFA 2021 Player Dataset (fifa21 raw data v2.csv).

Inside the notebook, the dataset is loaded from:

df = pd.read_csv('Fifa/fifa21 raw data v2.csv')

To successfully run the notebook, place the CSV file in a folder named:

Fifa/

Or update the file path in the notebook.

---

🧰 Libraries & Tools

The project uses several widely-adopted Python libraries:

• Pandas, NumPy
• Matplotlib, Seaborn, Plotly
• SciPy
• Scikit-learn
• XGBoost

---

🧹 Data Preprocessing & Cleaning

Text Cleaning

Contract Parsing

Height & Weight Standardization

Monetary Value Normalization

Star Ratings Parsing

---

🕳️ Handling Missing & Duplicate Data

• Logical imputation for missing values
• Handling "nan" text values correctly
• Detecting full and partial duplicates

---

📉 Outlier Detection

• Z-Score
• IQR
• Local Outlier Factor (LOF) on financial attributes

---

🔄 Transformations & Scaling

• Log Transform
• Box-Cox
• Yeo-Johnson
• Standard/MinMax/Robust Scaling

---

🧬 Feature Engineering & Encoding

• Frequency Encoding
• One-Hot Encoding
• Label & Ordinal Encoding

---

🎯 Feature Selection

• Correlation
• Mutual Information
• RFE / RFECV Using RandomForest-based ranking

---

🤖 Model Training: XGBoost Regressor

The model is trained to predict each player's Overall Rating after preprocessing and feature selection.

Evaluation metrics:

• R² Score
• MSE

---

🧪 Prediction Example

A complete example is provided in the notebook showing how to predict a new player's rating using transformed inputs.

---

▶️ Running Locally

pip install pandas numpy matplotlib seaborn plotly scipy scikit-learn xgboost
jupyter notebook FIFA2021_Comprehensive_EDA_&_Data_cleaning.ipynb

---

🎓 Educational Purpose

This project is intended as a comprehensive learning resource for:

• Data cleaning
• Data transformation
• Feature encoding
• Outlier analysis
• Feature selection
• Model training

---

✅ Contributions Welcome!

Feel free to fork, improve, and explore further.