Customer Purchase Behavior Analysis & Prediction for Amazon
Problem Statement:
Amazon, a global leader in e-commerce, wants to optimize its customer segmentation,
revenue forecasting, and churn prediction to enhance customer retention and increase
revenue. With millions of customers and transactions daily, Amazon collects demographic
details, purchase history, and transaction data but faces the following challenges:
Identifying high-value customers for targeted marketing.
Predicting Customer Lifetime Value (CLV) to improve revenue forecasting.
Understanding customer churn risks and improving retention strategies.
Grouping customers into actionable segments based on behavior patterns.
The goal of this project is to develop Machine Learning models to segment customers,
predict their future spending, and classify them as potential churners or active customers.
However, before building ML models, we need to clean and preprocess the data to ensure
accuracy.
Step 1: Data Cleansing & Preprocessing
Before applying ML models, it is crucial to ensure data quality by performing the following
steps:
Handling Missing Values
● Identify missing values in Age, Purchase Amount, Rating, and Customer Lifetime
Value (CLV).
● Apply mean/median imputation for numerical fields.
● Apply mode imputation for categorical fields like Payment Method.
Removing Duplicates
● Remove duplicate entries based on Customer_ID and Purchase_Date.
Data Formatting & Type Correction
● Convert Purchase_Date to datetime format.
● Standardize categorical values (e.g., Gender: Male, Female, Other).
● Ensure consistent data types (integers for numeric fields, categorical encoding for
non-numeric).
Handling Outliers
● Identify outliers in Purchase Amount & CLV using boxplots & z-score analysis.
● Apply winsorization or remove extreme outliers.
Feature Engineering (Adding New Columns)
To make the dataset more useful for machine learning, we add the following new columns:
1. Customer_Lifetime_Value (CLV): Projected future revenue per customer.
2. Loyalty Score: Score based on purchase frequency and total spending.
3. Discount Applied: Whether the purchase was made with a discount (Yes/No).
4. Return Status: Indicates if the item was returned (Yes/No).
5. Customer Segment: Categorized as New, Regular, VIP based on loyalty.
6. Preferred Shopping Channel: Where the customer shops (Online, In-store, Both).
Step 2: Machine Learning Tasks
After data cleaning and feature engineering, we apply Machine Learning models to derive
insights.
Customer Segmentation (Clustering - K-Means)
Objective:
● Categorize Amazon customers into distinct groups based on spending patterns,
purchase frequency, and loyalty scores.
● Identify high-value, occasional, and low-value customers for targeted promotions.
Method:
● Use K-Means Clustering to segment customers into groups based on:
o Total purchase amount
o Number of orders
o Loyalty score
Industry Application:
● Helps Amazon personalize recommendations and promotions for different
customer segments.
● Enables dynamic pricing strategies based on customer type.
Predicting Customer Lifetime Value (Regression - Linear Regression)
Objective:
● Estimate the future revenue Amazon can generate from each customer.
● Identify high-CLV customers and offer exclusive deals to increase retention.
Method:
● Train a Linear Regression model to predict CLV based on:
o Age, past purchases, discount usage, payment method, and loyalty score.
Industry Application:
● Helps Amazon in predictive marketing and resource allocation.
● Enables cost-efficient retention strategies.
Expected Deliverables
✔ Cleaned dataset with new features (CLV, Loyalty Score, etc.).
✔ K-Means Clustering for customer segmentation.
✔ Linear Regression model for CLV prediction.
✔ Logistic Regression model for churn prediction.
✔ Power BI dashboard for visualizing insights.
✔ Jupyter Notebook with all models & findings.
