pca1 vs pca2 to observe cluster of customers. Same coloured dots represent customers put in common cluster
Studying customer behaviour is key to market study. Individuals show different brand loyalty and degree of satisfaction towards same product or service. And why won't they, annual income, bank services, social status & living standards are not same for all. Hence, it is of utmost importance to target different groups with different marketting strategy. This project aims to study the same by segmentting a credit card company's customer base into four major clusters.
- Dataset: Credit card details of ~ 9k customers
- Features: Balance, Purchase, Purchase frequency, Credit Limit, Minimum Payment, Tenure and 11 others
- Target: Classify customers into groups to observe their spending behaviour
- Alienated - Low Balance, Low Purchase - 6186 - RED
- Roamers - Medium to high Balance, Low Purchase - 1330 - ORANGE
- Supporters - Low Balance, High Purchase - 1324 - GREEN
- Fans - High purchase - 110 - DARK GREEN
- Libraries: KMeans, PCA, StandardScaler, tensorflow, numpy, pandas
- Algorithm: KMeans Clustering
- Activation Function|Optimizer: ReLU|Adam
- No. of clusters: 4
- Dimentionality Reduction: Autoencoder (TensorFlow library), PCA
- Insights:
- Most customers tend to do payment in installments, as full payment percentage is only 15%
- A large group of customers show recent purchase in range 1,000-3,000. Only a handful of them have done purchases as high as 50,000
- One Off Purchase and Cash Advance frequency is LOW
- Tenure shows maximum correation with Credit limit
- Follow below link for complete step-by-step analysis
Github Link
Future Scope
This same model can be used to classify and study customers of any store. All I need is your store's data.