Project Details

BCG Customer Churn Prediction

Project Overview

The BCG Customer Churn Prediction project focuses on predicting high-risk SME customers for PowerCo using machine learning. By integrating customer demographics, usage metrics, and engagement history, the project provides actionable insights for targeted retention strategies. The solution leverages Random Forest, with additional preprocessing and feature engineering, to ensure accurate and reliable predictions.


Want to explore the full interactive notebook? Click here to view it on Kaggle.


BCG Customer Churn Certificate

Situation

PowerCo faces challenges with customer retention, particularly among SME clients. Churned customers lead to revenue loss and increased acquisition costs. Existing monitoring processes are reactive, relying on historical reporting rather than predictive insights, making it difficult to identify high-risk customers proactively.

Task

The objective was to develop a predictive model that enables PowerCo to:

  • Identify high-risk SME customers likely to churn.
  • Understand the key factors influencing churn.
  • Provide actionable insights to guide retention campaigns.
  • Improve decision-making through predictive analytics rather than reactive reporting.

Action

The project was executed through a structured machine learning pipeline:

  • Data Preprocessing

    • Handled missing values and outliers.
    • Encoded categorical features and scaled numerical data.
    • Applied Yeo-Johnson and log transformations to normalize skewed variables.

  • Feature Engineering

    • Created additional metrics capturing customer engagement and usage patterns.
    • Selected features most predictive of churn using correlation and importance analysis.

  • Modeling

    • Trained a Random Forest classifier to predict customer churn.
    • Evaluated performance using metrics such as accuracy (~90%) and precision (~0.8).
    • Fine-tuned hyperparameters to optimize model performance.

Result

The BCG Customer Churn Prediction project delivers:

  • Accurate identification of high-risk SME customers, enabling proactive retention.
  • Insights into key churn drivers, supporting strategic decision-making.
  • A reliable machine learning workflow, from preprocessing to prediction.
  • Potential for targeted campaigns that reduce churn and increase revenue.

By combining rigorous data preprocessing, feature engineering, and Random Forest modeling, this solution equips PowerCo with a predictive analytics framework to minimize churn and improve customer retention strategies.