Customer Analysis and Segmentation Using K-Means

Overview

This project involves analyzing and segmenting customers of a UK-based online giftware store. The analysis focuses on feature engineering, K-Means clustering, and visualization techniques to derive insights into customer behavior.

Technologies Used

• Python
• Pandas
• Matplotlib
• Seaborn
• Scikit-learn

Data Source

The dataset for this analysis can be found at: UCI Machine Learning Repository - Online Retail II.

Project Structure

• Data Exploration: Initial exploration of the dataset, including checking for missing values and data types.
• Data Cleaning: Processes to clean the dataset by removing invalid entries and handling missing values.
• Feature Engineering: Creating new features that help in understanding customer behavior.
• K-Means Clustering: Implementing K-Means clustering to segment customers based on their purchasing patterns.
• Visualization: Creating visual representations of the clustered data for better understanding.

Analysis Steps

Data Exploration

• Loaded the dataset and performed preliminary checks.
• Noted the presence of null values in the Customer ID column and negative values in the Quantity and Price columns.
• Observed that some invoices and stock codes did not conform to expected formats.

Data Cleaning

• Removed rows with null Customer ID values and negative quantities.
• Dropped transactions with zero or negative prices.
• Filtered the dataset to retain only valid invoices (6-digit numbers).

Feature Engineering

• Calculated SalesLineTotal as the product of Quantity and Price.
• Aggregated data by Customer ID to compute:
  • Monetary Value: Total spending per customer.
  • Frequency: Number of unique invoices per customer.
  • Last Invoice Date: The most recent transaction date per customer.

K-Means Clustering

• Applied K-Means clustering using the features of Monetary Value, Frequency, and Recency (derived from Last Invoice Date).
• Evaluated the clustering results using silhouette scores to determine the optimal number of clusters.

Visualization

• Created visualizations to depict clusters and customer segments.
• Utilized seaborn and matplotlib for effective data representation.

Notes

• The dataset had significant missing values in the Customer ID column, which required careful handling.
• Negative quantities and prices indicated potential data entry errors or cancellations, which were filtered out during data cleaning.
• The analysis revealed distinct customer segments that could be targeted with tailored marketing strategies.

Conclusion

The project successfully segmented customers of the giftware store, providing valuable insights into purchasing behavior. These insights can guide marketing efforts and improve customer relationship management.

Future Work

• Further refining the clustering algorithm to enhance segmentation accuracy.
• Incorporating additional features like customer demographics for deeper analysis.
• Developing a predictive model to forecast future customer behavior.

Acknowledgements

Thanks to the UCI Machine Learning Repository for providing the dataset used in this analysis.