Using K-Means Clustering Algorithm for Mall Customer Segmentation: From Elbow Method to Visualization Practice

This article presents a complete machine learning practical project that uses the K-Means clustering algorithm to segment mall customers. The core process includes: using the elbow method to determine the optimal number of clusters K=5, visually displaying the feature distribution of five customer groups, and helping to understand the practical application of unsupervised learning in business analysis. This project is derived from the SkillCraft Machine Learning Internship Task, and the original project was published by srethulak on GitHub (link: https://github.com/srethulak/SkillCraft-ML-Task02-Mall-Customer-Segmentation).

Customer segmentation is a classic application scenario of unsupervised learning, which can discover hidden patterns in data without labeled data. For retail businesses, understanding the characteristics of different customer groups is key to formulating precise marketing strategies. This project fully demonstrates the entire process from data loading, feature selection, model training to result visualization, and uses K-Means to segment mall customers by annual income and spending score, providing support for differentiated marketing.

The project uses the classic Mall Customer Dataset and selects two key features:

• Annual Income: in thousands of US dollars, reflecting purchasing power;
• Spending Score: 1-100 points, reflecting consumption willingness. The combination of these two can effectively distinguish customer groups (ability to buy vs. willingness to buy).

K-Means requires pre-specifying the K value, and the elbow method finds the optimal K through WCSS (Within-Cluster Sum of Squares):

1. Principle: WCSS decreases as K increases, and the inflection point indicates the optimal K;
2. Implementation: Iterate K from 1 to 10, initialize with k-means++, and record WCSS;
3. Conclusion: When K=5, the curve shows an obvious elbow, which is the optimal number of clusters.

The clustering results when K=5 are divided into 5 groups:

• Low Income & Low Spending Group: Low income and low consumption willingness;
• Low Income & High Spending Group: Not high income but strong consumption willingness;
• Medium Income & Medium Spending Group: Medium income and medium consumption;
• High Income & Low Spending Group: High income but conservative consumption;
• High Income & High Spending Group: High income and active consumption (core value customers). Differentiated marketing strategies can be formulated for different groups, such as pushing high-end promotions to the High Income & Low Spending Group.

The technical process includes:

1. Data Preprocessing: Load data with pandas and select the annual income and spending score columns;
2. Model Training: Use scikit-learn's KMeans with parameters n_clusters=5, init='k-means++', random_state=42;
3. Visualization: Use matplotlib to draw scatter plots, distinguish clusters with different colors, and mark cluster centers.

Insights from the project:

• Feature Selection: More dimensions (such as age, gender) are needed in real scenarios;
• Business Interpretation: Clustering results need to be combined with domain knowledge to transform into strategies;
• Algorithm Limitations: K-Means assumes spherical clusters and does not perform well on non-convex data or data with large density differences; algorithms like DBSCAN can be tried.

This project fully demonstrates the application of K-Means in customer segmentation, from K value determination to visualization, making it a good practice project for getting started with unsupervised learning. The value of customer segmentation lies in transforming data insights into business actions—technology provides possibilities, and business creates a closed loop of value.