Churn Modelling — Data Preprocessing Pipeline

A step-by-step machine learning preprocessing pipeline built on the Bank Churn Modelling dataset. This project covers the full journey from raw, messy data to a clean, encoded, and scaled dataset ready for model training.

---

📋 Project Overview

The goal is to preprocess customer banking data to predict churn (whether a customer will leave the bank). Each notebook in this project handles one stage of the preprocessing pipeline:

Step	Notebook	Description
0	0_handle_missing_values.ipynb	Impute missing ages with mean/median; predict missing genders using an LLM (Groq API)
1	1_handling_outliers.ipynb	Detect outliers using KDE plots, box plots, 3-sigma rule, and IQR method
2	2_feature_binning.ipynb	Bin CreditScore into ordinal categories (Poor → Excellent)
3	3_feature_encoding.ipynb	One-hot encode nominal columns (Geography, Gender); label encode ordinal column (CreditScoreBins)
4	4_feature_scaling.ipynb	Standardize continuous features (Age, Tenure, Balance, EstimatedSalary) using Z-score scaling

Data Flow

ChurnModelling.csv  (raw)
        ↓  0_handle_missing_values
ChurnModelling_Missing_values_handled.csv
        ↓  1_handling_outliers
ChurnModelling_Outliers_Handled.csv
        ↓  2_feature_binning
ChurnModelling_Binning_Applied.csv
        ↓  3_feature_encoding
ChurnModelling_Encoded.csv
        ↓  4_feature_scaling
ChurnModelling_Final.csv  ✅ model-ready

---

🐍 What is Anaconda?

Anaconda is a free distribution of Python that comes pre-packaged with hundreds of data science libraries (NumPy, pandas, scikit-learn, etc.) and the conda package manager. Instead of manually installing Python and every library one by one, Anaconda gives you everything in one installer.

Think of it like this:

• Python alone = a kitchen with no equipment
• Anaconda = a fully equipped kitchen, ready to cook

Download it here: https://www.anaconda.com/download

---

📦 What is a Virtual Environment?

A virtual environment is an isolated Python workspace. It has its own Python version and its own set of installed packages, completely separate from every other project on your machine.

Why does this matter?

Imagine two projects:

• Project A needs pandas==1.5
• Project B needs pandas==3.0

Without environments, installing one would break the other. With environments, each project lives in its own bubble and never interferes with anything else.

Your computer
├── env_project_a/    ← pandas 1.5, scikit-learn 1.0
├── env_project_b/    ← pandas 3.0, scikit-learn 1.8
└── env_this_project/ ← exact versions in requirements.txt

---

🚀 Setup Instructions

1. Clone the repository

git clone https://github.com/your-username/your-repo-name.git
cd your-repo-name

2. Create a conda environment

conda create -n bprmls python=3.11

This creates a fresh isolated environment named bprmls running Python 3.11.

3. Activate the environment

conda activate bprmls

Your terminal prompt will change to show (bprmls) — this means you are now inside the environment. Any packages you install from this point go into this environment only.

4. Install dependencies

pip install -r requirments.txt

5. Set up your .env file (required for notebook 0)

Notebook 0_handle_missing_values.ipynb uses the Groq API to predict missing gender values from customer names. You need a free API key from https://console.groq.com.

Create a file named .env in the root of the project:

GROQ_API_KEY=your_api_key_here

⚠️ Never commit your .env file to GitHub. It is already listed in .gitignore.

6. Place the raw data

Make sure the raw dataset is placed at:

data/raw/ChurnModelling.csv

The data/processed/ folder will be populated automatically as you run the notebooks in order.

7. Open Jupyter and run notebooks in order

jupyter notebook

Run notebooks 0 → 1 → 2 → 3 → 4 in sequence. Each notebook reads the output CSV from the previous step.

---

📁 Project Structure

├── data/
│   ├── raw/
│   │   └── ChurnModelling.csv
│   └── processed/
│       ├── ChurnModelling_Missing_values_handled.csv
│       ├── ChurnModelling_Outliers_Handled.csv
│       ├── ChurnModelling_Binning_Applied.csv
│       ├── ChurnModelling_Encoded.csv
│       └── ChurnModelling_Final.csv
├── 0_handle_missing_values.ipynb
├── 1_handling_outliers.ipynb
├── 2_feature_binning.ipynb
├── 3_feature_encoding.ipynb
├── 4_feature_scaling.ipynb
├── requirments.txt
├── .env                  ← you create this (not committed)
└── README.md

---

📚 Techniques Covered

Missing Value Handling

• Row/column deletion
• Mean & median imputation for numerical features
• LLM-based imputation for categorical features (Groq API)

Outlier Detection

• KDE distribution plots
• Box plots
• 3-Sigma (empirical) rule
• IQR method

Feature Binning

• Custom binning of CreditScore into interpretable categories: Poor / Fair / Good / Very Good / Excellent

Feature Encoding

• One-hot encoding for nominal variables (Geography, Gender) using pd.get_dummies
• Ordinal / label encoding for CreditScoreBins using a manual mapping dictionary
• sklearn OneHotEncoder and LabelEncoder as an alternative approach

Feature Scaling

• Z-score standardization using sklearn.preprocessing.StandardScaler on Age, Tenure, Balance, EstimatedSalary

---

🔧 Requirements

See requirments.txt for the full list. Key libraries:

• numpy
• pandas
• scikit-learn
• seaborn
• matplotlib
• openai
• groq
• python-dotenv

---

⚠️ Notes

• Notebooks must be run in order (0 → 4) since each reads the output of the previous one.
• The .env file is required only for notebook 0. All other notebooks work without an API key.
• The data/processed/ folder is gitignored. Run the notebooks to regenerate the processed files.