🚀 Live Demo: Churn & Prediction App
This project leverages machine learning to predict customer churn in the banking sector. By identifying at-risk customers, we aim to help banking professionals take proactive measures, retain customers, and improve long-term customer relationships. The model not only predicts churn but also provides insights into customer risk factors and can be integrated into personalized retention strategies, such as tailored incentives.
- Churn Prediction: Uses advanced ML models to predict the likelihood of customer churn.
- Risk Factor Analysis: Highlights reasons for each customer’s risk of attrition.
- Personalized Retention Campaigns: Generates custom email incentives based on customer risk factors.
- Fraud Detection: Employs ML models to flag potential fraudulent transactions.
_ Table of Contents _
- Data Preprocessing
- Model Training and Evaluation
- Acknowledgments
Project Structure
This project uses a dataset from Kaggle (linked in Resources) that contains 14 features. Key preprocessing steps include:
- Feature Selection: Dropping irrelevant columns such as CustomerId and Surname. Encoding Categorical Variables: Converting Geography and Gender columns to numerical using one-hot encoding.
- Data Scaling: Standardizing features to improve model performance.
The following models were trained and evaluated on the processed dataset:
- XGBoost
- Gradient Boosting
- Random Forest
- Decision Tree
- Naive Bayes
- Support Vector Classifier (SVC)
Each model was evaluated using accuracy, precision, recall, and F1-score, with a focus on recall to capture at-risk customers.
Here are the performances for each of our models:
For Fraud detection here how the best version of the model performed
Given the imbalance in churn vs. non-churn customers, SMOTE (Synthetic Minority Oversampling Technique) was used to balance the dataset, enhancing recall performance.
Thanks to the Headstarter team for their guidance and feedback.A special thanks to Faizan (Co-founder) for his support and mentorship.
Dataset: Kaggle - Churn for Bank Customers Preprocessing insights: LakeFS Blog on Data Preprocessing