Customer Churn Prediction

Overview

This project focuses on predicting customer churn using machine learning techniques. The goal is to identify customers who are likely to stop using a company's services, enabling businesses to take proactive measures to retain them.

A Support Vector Machine(SVM) classifier was used to build the model, which was then deployed using a Flask web application with API endpoints for real-time predictions.

Project Structure

• Exploratory Data Analysis (EDA): Analysis to identify key features that affect customer churn.
• Model Building: To improve model performance, multiple approaches, including undersampling, oversampling using SMOTE and hyperparameter tuning, were tried.
• Deployment: The final model was deployed using Flask with a functional API.

Key Features

• Model: Random Forest Classifier

• Performance:

  • Accuracy: 87%
  • Recall: 49%

• Model: Decision Tree (After undersampling)

• Performance:

  • Accuracy: 82%
  • Recall: 60%

  Model: Support Vector Machine (After SMOTE)

• Performance:

  • Accuracy: 84%
  • Recall: 68%

• Deployment: Flask app with API endpoints, tested with Postman.

Methods Used

• Data Preprocessing: Handled missing values, removed outliers, and scaled features using StandardScaler.
• Feature Engineering: Experimented with new features but avoided methods that reduced model performance.
• Undersampling: Applied to improve model balance and recall.
• SMOTE: Applied to improve the recall and solve the class imbalance problem.

Getting Started

Requirements

• Python 3.x
• Libraries:
  • Flask
  • Scikit-learn
  • Pandas
  • Numpy
  • Postman (for testing API)

Installation

1. Clone the repository:

   git clone https://github.com/jayjoshi33/customer-churn-prediction.git

2. Install dependencies:

   pip install -r requirements.txt

3. Run the Flask app:

   python app.py

4. Use Postman or a similar tool to test the API by sending JSON input for predictions.

API Usage

• Endpoint: /predict
• Method: POST
• Input: Customer data in JSON format.
• Output: Predicted class (churn or not churn).

Results

• The Random Forest model achieved 87% accuracy with 49% recall.
• After applying SMOTE, a SVC model improved recall to 68%.

Conclusion

This project provides a predictive model for customer churn, allowing businesses to identify potential churners and take necessary actions. The model is deployed via Flask, with easy API access for real-time predictions.