Credit Card Segmentation - Data Science Capstone Project

Overview

This project focuses on customer segmentation to define a targeted marketing strategy based on the usage behavior of approximately 9000 active credit card holders over the last 6 months. The dataset includes customer-level information with 18 behavioral variables.

Files

• credit-card-data: Raw data containing customer information.
• Project_python1.ipynb: Jupyter notebook with Python code.
• Project_R_1.R: R code.
• Project Report.doc: Detailed project report with visualizations and behavioral patterns.
• Credit-Card Segmentation Project DF_Python.csv: Cluster results formed with Python code.
• Credit-Card Segmentation Project DF_R.csv: Cluster results formed with R code.

Characteristics Analysis

Cluster 1: Good Payers, Hesitant Users

• Not frequent credit card users.
• Majority use cash advance transactions with low frequency.
• Low credit limit with a high payment ratio.
• Low usage limit and percentage for full payment.

Marketing Strategy Ideas:

• Extend credit limits.
• Reduce transaction charges for cash advances.
• Provide low-interest rates and incentives for purchases.

Cluster 2: Good Payers and Good Users

• Good cash advance and decent purchase type users.
• Good payment ratio and credit usage.
• Slightly low percentage of full payment.

Marketing Strategy Ideas:

• Offer high reward points.
• Eliminate late fees and minor charges.
• Provide internationally accepted cards.

Cluster 3: Mediocre Users with Some Defaulters

• Good purchase type users.
• Poor cash advance users.
• Some defaulters with high payment ratio and low credit limit.

Marketing Strategy Ideas:

• Offer reduced interest rates for expensive products.
• Advertise advantages of using cash advance transactions.

Data Analysis

1. Missing Value Analysis

• Missing values in MINIMUM_PAYMENTS: 313
• Imputed using KNN imputation method.

2. Feature Engineering

• New variable Card_use_type derived to indicate the type of transaction.
• Monthly average purchase and cash advance amount calculated.
• Limit usage, payment ratio, and other features engineered.

3. Outlier Analysis

• Outliers dropped for variables: BALANCE, CREDIT_LIMIT, PAYMENTS.

4. Correlation Analysis

• Heatmap used to identify and drop correlated variables.

5. Normalization

• Numerical data normalized using standardization technique.

6. Model Development

• Elbow method used to determine the optimum number of clusters (3).
• K-means clustering applied, and observations classified into three clusters.

Cluster Counts

• Cluster 1: 3340
• Cluster 2: 2701
• Cluster 3: 1239

For more details, please refer to the Project Report.