Customer Churn Classification Project

This is my final project from a machine learning course, where I applied everything I learned by building a real-world churn prediction system.

---

Project Overview

This project predicts whether a bank customer is likely to churn (leave the bank) based on personal and account data. I approached the problem using both classical ML models and a neural network (ANN), while optimizing performance and usability.

---

Project Workflow

1. Data Preprocessing

• Cleaned and prepared a customer dataset
• Handled feature scaling, encoding, and class imbalance
• Applied proper transformations to match model inputs (especially for Streamlit)

2. Model Selection with PyCaret

• Used pycaret.classification to:

  • Automate model comparison
  • Rank classifiers (Random Forest, SVM, XGBoost, ANN, e.g.)
  • Identify top 3 models based on accuracy and AUC
  • Exported hyperparameters from PyCaret for custom model building

3. Custom Modeling

• Trained Random Forest using parameters optimized by PyCaret

• Trained a Deep Learning ANN using TensorFlow/Keras

• Tracked metrics including accuracy, F1-score, precision, recall, and AUC

• Prevented overfitting using:

  • Class weights
  • Early stopping
  • Validation curves
  • ROC analysis

4. Web App Interface

• Built a simple and interactive Streamlit app for customer input
• Used the trained Random Forest model to predict churn
• Applied live preprocessing (one-hot encoding, scaling) before prediction
• Added UX feedback (probabilities, result summary)

---

Platforms Used :

• Google Colab: For training heavy models (PyCaret, ANN)
• Local Machine: For model testing, tuning, and Streamlit frontend
• Conda & Virtualenv for environment isolation

• Requirements

---

I’ve included:

• requirements.txt: For all packages used on local hardware
• .pkl or .h5 model files: For deployment
• Trained scaler.pkl: To match preprocessing with training

---

How to Run the App

pip install -r requirements.txt
streamlit run streamlit_app.py

---

Files Overview

File/Folder	Description
churn_preprocessed.csv	Cleaned dataset
model.pkl / ann_model.h5	Trained model (RF or ANN)
scaler.pkl	Fitted scaler for input standardization
streamlit_app.py	Web app UI using Streamlit
requirements.txt	Local environment dependencies

---

Hossein Final project for ML course | Practicing real-world deployment