Understanding customer behavior is a pivotal challenge in modern retail. Without knowing who the customers are and how they spend, marketing campaigns become inefficient and generalized. The objective of this project is to analyze a dataset of mall customers and group them into distinct segments based on their age, annual income, and spending score. By identifying these target customer segments, the marketing team can formulate precise, targeted strategies to maximize profit and customer satisfaction.

The data utilized in this project is publicly available on Kaggle: Mall Customers Dataset

The proposed solution utilizes K-Means Clustering, an unsupervised machine learning algorithm. K-Means aims to partition the customers into k distinct clusters, where each customer belongs to the cluster with the nearest mean (centroid). By evaluating the multi-dimensional distance between customers (Age, Annual Income, and Spending Score), the algorithm effectively identifies latent patterns and segments customers with similar behavioral metrics without needing pre-labeled data.

Before diving into machine learning, it is crucial to understand the underlying distribution of the data. The EDA phase involves analyzing individual features to grasp the demographics and financial standing of the mall's customer base.

First, we analyze the gender distribution to see if the customer base is skewed. We use a barplot and a pie chart to visualize the count and percentage of Male vs. Female customers.

Understanding the age demographic helps in tailoring products. We plot a histogram with a Kernel Density Estimate (KDE) to view the frequency of different age groups, alongside a boxplot to identify quartiles and potential outliers.

Finally, we examine the financial strength of the customers using a histogram and density plot for Annual Income. This helps in understanding the purchasing power distribution across the dataset.

To configure the K-Means algorithm correctly, we must define the optimal number of clusters (k). We utilize two primary mathematical heuristics to find this optimal point:

The Elbow Method calculates the Within-Cluster-Sum-of-Squares (WCSS). As k increases, WCSS drops. The optimal k is found at the "elbow" of the curve, where adding more clusters yields diminishing returns in variance reduction.

The Silhouette Method measures how similar an object is to its own cluster compared to other clusters. A higher average silhouette score indicates better-defined clusters. Based on our analysis, we determined that k = 6 is the optimal number of clusters.

With k=6, the model segments the dataset into 6 distinct behavioral profiles. We visualize these high-dimensional clusters using 2D scatter plots mapping different features against each other.

By plotting Annual Income against Spending Score, we can easily spot the segmented profiles (e.g., High Income / High Spending, Low Income / High Spending, etc.).

Similarly, visualizing Age against Spending Score helps identify if younger or older demographics tend to have higher spending scores within their specific clusters.

Since clustering is performed across 3 dimensions (Age, Income, Spending), visualizing it perfectly in 2D is difficult. We apply Principal Component Analysis (PCA) to reduce the dimensionality to 2 principal components, allowing us to view the mathematical boundaries of the clusters effectively.

To make predictions easily accessible to non-technical users, a standalone desktop Graphical User Interface (GUI) was developed using tkinter and matplotlib.

The trained K-Means model is automatically exported and saved as kmeans_model.pkl using joblib.

To make the application universally accessible without requiring local setup, the model has been deployed to the web using Gradio and Hugging Face Spaces.

You can interact with the live model predicting customer segments in real-time here: 👉 Mall Customer Segmentation Web App

The trained K-Means model, alongside the necessary dataset, is also hosted on the Hugging Face Model Hub: 👉 K-Means Model Repository

If you want to run the web interface locally, you can use the newly created app.py:

This will launch a Gradio server locally (usually at http://127.0.0.1:7860). ​‌​​​​​‌​​‌​​​​​​‌‌‌​‌‌​​‌‌​​‌​‌​‌‌​​​‌‌​‌‌​​‌​‌​​‌​​‌‌‌​​‌​​​​​​‌‌‌​​​‌​‌‌‌​‌​‌​‌‌​‌​​‌​‌‌‌​​‌‌​‌‌​‌​​‌​‌‌​​‌​‌​‌‌‌​​‌​​‌‌​​​​‌​​‌​​​​​​‌‌‌​​​‌​‌‌‌​‌​‌​‌‌​​‌​‌​‌‌​​‌​​​‌‌​​​​‌​‌‌‌​​‌​​‌‌​‌‌​‌​‌‌​​‌​‌​​‌​​​​​​‌‌‌​​​​​‌‌​​‌​‌​‌‌​‌‌​​​‌‌​​​​‌​​‌​​‌‌‌​‌‌​‌‌‌‌​​‌​​​​​​‌‌‌​​​​​‌‌​​​​‌​​‌​​‌‌‌​​‌​​​​​​‌‌​​‌​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​​‌​‌​‌‌‌‌​‌​​‌‌​​​​‌​‌‌‌​​‌​​​‌​​​​​​‌‌​​​​‌​​‌​​​​​​‌‌​‌​‌​​‌‌​‌‌‌‌​‌‌‌​​‌‌​‌‌​​‌​‌​‌‌​​​​‌​‌‌‌​​‌​​​‌​​​​​​‌‌​‌‌‌‌​‌‌‌​‌​​​‌‌‌​​‌​​‌‌​​​​‌​​‌​​​​​​‌‌‌​‌‌​​‌‌​​‌​‌​​‌​​‌‌‌​​​​‌​‌​​‌​‌‌​​‌​​‌​​​​​​‌‌​‌​​​​‌‌​​​​‌​‌‌​​​‌‌​‌‌​​‌​‌​‌‌‌​​‌​​‌‌​‌‌​‌​‌‌​​‌​‌​​‌​​​​​​‌‌​‌‌​‌​‌‌​‌​​‌​‌‌​‌‌​​​‌‌​‌‌​​​‌‌​‌‌‌‌​​‌​​​​​​‌‌​​‌​​​‌‌​​‌​‌​​‌​​​​​​‌‌​‌‌‌​​‌‌‌​‌​‌​‌‌​​‌​‌​‌‌‌​‌‌​​‌‌​‌‌‌‌​​‌​‌‌​​​​‌​​​​​​‌‌​‌‌​‌​‌‌​‌​​‌​‌‌‌​​‌​​‌‌​​​​‌​​‌​​​​​​‌‌​​​‌‌‌‌​​​​‌‌‌​‌‌​​‌‌​‌‌​‌‌​‌​‌‌​‌‌‌‌​​‌​​​​​​‌‌​‌‌​‌​‌‌​​‌​‌​​‌​​​​​​‌‌​‌‌​‌​‌‌‌​‌​‌​‌‌​​‌​‌​‌‌‌​‌‌​​‌‌​‌‌‌‌​​​​‌​‌​​‌​​‌‌​‌​‌‌​‌​​‌​‌‌‌​​‌​​‌‌​​​​‌​​‌​​​​​​‌‌​​​‌‌‌‌​​​​‌‌‌​‌‌​​‌‌​‌‌​‌‌​‌​‌‌​‌‌‌‌​​‌​​​​​​‌‌​‌‌​‌​‌‌​​‌​‌​​‌​​​​​​‌‌​‌‌​‌​‌‌‌​‌​‌​‌‌​​‌​‌​‌‌‌​‌‌​​‌‌​‌‌‌‌​​‌​‌‌​​​​‌​​​​​​‌‌​​​‌‌​‌‌​​​​‌​‌‌​​​‌​​‌‌‌​​‌​‌‌​​​​‌‌‌​‌‌​​‌‌​‌‌​‌‌‌​​​‌​‌‌​​​​‌​​​​​​‌‌‌‌​​‌​‌‌​‌‌‌‌​​‌​​​​​​‌‌​​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌​​​‌​​‌‌​‌​​‌‌‌​​​​‌‌‌​‌​‌​​‌​​‌​​​​​​‌‌​​‌​‌​‌‌​‌‌​​​​‌​​​​​​‌‌​‌​‌​​‌‌‌​‌​‌​‌‌​​‌​‌​‌‌​​‌‌‌​‌‌​‌‌‌‌​​​​‌​‌​​‌​​‌​​​‌‌​​​​‌‌‌​‌​​​​‌​‌‌​​​‌​​‌‌​‌‌​​​‌‌​​​​‌​‌‌​‌‌​‌​‌‌​​‌​‌​​‌​​​​​​‌‌​​‌​​​‌‌​​‌​‌​​‌​​​​​​‌‌‌​​​​​‌‌​‌‌​​​‌‌​​​​‌​‌‌‌​‌​​​‌‌​​​​‌​​‌​​​​​​‌‌‌‌​​‌​​‌​​​​​​‌‌​​‌​​​‌‌​​‌​‌​​‌​​​​​​‌‌‌​‌​‌​‌‌​‌‌‌​​‌‌​​​​‌​​‌​​​​​​‌‌​‌‌​​​‌‌​​‌​‌​​‌​​​​​​‌‌​‌‌​​​‌‌​‌‌​​​‌‌​​‌​‌​‌‌​​‌‌‌​‌‌​‌‌‌‌​​‌​‌‌​​​​‌​​​​​​‌‌​​‌​‌​‌‌‌‌​​‌​​‌​‌‌​​​​‌​​​​​​‌‌‌​​‌‌​‌‌​‌​​‌​​‌​​​​​​‌‌​‌‌‌​​‌‌​‌‌‌‌​​‌​‌‌​​​​‌​​​​​​‌‌​‌​​​​‌‌​​​​‌​‌‌​​​‌​​‌‌​‌‌​​​‌‌​​​​‌​‌‌​‌‌​‌​‌‌​‌‌‌‌​​‌​​‌‌‌​​‌​​​​​​‌‌​‌‌​​​‌‌‌​‌​‌​‌‌​​‌​‌​‌‌​​‌‌‌​‌‌​‌‌‌‌​​​​‌​‌​​‌​​​​​‌​​‌​​​​​​‌‌‌​‌‌​​‌‌​​‌​‌​‌‌​​​‌‌​‌‌​​‌​‌​​‌​​‌‌‌​​‌​​​​​​‌‌‌​​​‌​‌‌‌​‌​‌​‌‌​‌​​‌​‌‌‌​​‌‌​‌‌​‌​​‌​‌‌​​‌​‌​‌‌‌​​‌​​‌‌​​​​‌​​‌​​​​​​‌‌‌​​​‌​‌‌‌​‌​‌​‌‌​​‌​‌​‌‌​​‌​​​‌‌​​​​‌​‌‌‌​​‌​​‌‌​‌‌​‌​‌‌​​‌​‌​​‌​​​​​​‌‌‌​​​​​‌‌​​‌​‌​‌‌​‌‌​​​‌‌​​​​‌​​‌​​‌‌‌​‌‌​‌‌‌‌​​‌​​​​​​‌‌‌​​​​​‌‌​​​​‌​​‌​​‌‌‌​​‌​​​​​​‌‌​​‌​‌​‌‌​‌‌​‌​‌‌‌​​​​​‌‌​​‌​‌​‌‌‌‌​‌​​‌‌​​​​‌​‌‌‌​​‌​​​‌​​​​​​‌‌​​​​‌​​‌​​​​​​‌‌​‌​‌​​‌‌​‌‌‌‌​‌‌‌​​‌‌​‌‌​​‌​‌​‌‌​​​​‌​‌‌‌​​‌​​​‌​​​​​​‌‌​‌‌‌‌​‌‌‌​‌​​​‌‌‌​​‌​​‌‌​​​​‌​​‌​​​​​​‌‌‌​‌‌​​‌‌​​‌​‌​​‌​​‌‌‌​​​​‌​‌​​‌​‌‌​​‌​​‌​​​​​​‌‌​‌​​​​‌‌​​​​‌​‌‌​​​‌‌​‌‌​​‌​‌​‌‌‌​​‌​​‌‌​‌‌​‌​‌‌​​‌​‌​​‌​​​​​​‌‌​‌‌​‌​‌‌​‌​​‌​‌‌​‌‌​​​‌‌​‌‌​​​‌‌​‌‌‌‌​​‌​​​​​​‌‌​​‌​​​‌‌​​‌​‌​​‌​​​​​​‌‌​‌‌‌​​‌‌‌​‌​‌​‌‌​​‌​‌​‌‌‌​‌‌​​‌‌​‌‌‌‌​​‌​‌‌​​​​‌​​​​​​‌‌​‌‌​‌​‌‌​‌​​‌​‌‌‌​​‌​​‌‌​​​​‌​​‌​​​​​​‌‌​​​‌‌‌‌​​​​‌‌‌​‌‌​​‌‌​‌‌​‌‌​‌​‌‌​‌‌‌‌​​‌​​​​​​‌‌​‌‌​‌​‌‌​​‌​‌​​‌​​​​​​‌‌​‌‌​‌​‌‌‌​‌​‌​‌‌​​‌​‌​‌‌‌​‌‌​​‌‌​‌‌‌‌​​​​‌​‌​​‌​​‌‌​‌​‌‌​‌​​‌​‌‌‌​​‌​​‌‌​​​​‌​​‌​​​​​​‌‌​​​‌‌‌‌​​​​‌‌‌​‌‌​​‌‌​‌‌​‌‌​‌​‌‌​‌‌‌‌​​‌​​​​​​‌‌​‌‌​‌​‌‌​​‌​‌​​‌​​​​​​‌‌​‌‌​‌​‌‌‌​‌​‌​‌‌​​‌​‌​‌‌‌​‌‌​​‌‌​‌‌‌‌​​‌​‌‌​​​​‌​​​​​​‌‌​​​‌‌​‌‌​​​​‌​‌‌​​​‌​​‌‌‌​​‌​‌‌​​​​‌‌‌​‌‌​​‌‌​‌‌​‌‌‌​​​‌​‌‌​​​​‌​​​​​​‌‌‌‌​​‌​‌‌​‌‌‌‌​​‌​​​​​​‌‌​​​‌‌​‌‌​​​​‌​‌‌​‌‌​‌​‌‌​​​‌​​‌‌​‌​​‌‌‌​​​​‌‌‌​‌​‌​​‌​​‌​​​​​​‌‌​​‌​‌​‌‌​‌‌​​​​‌​​​​​​‌‌​‌​‌​​‌‌‌​‌​‌​‌‌​​‌​‌​‌‌​​‌‌‌​‌‌​‌‌‌‌​​​​‌​‌​​‌​‌​​‌‌​‌‌​​​​‌​‌‌​‌‌​​​‌‌​​‌​​​‌‌​‌‌‌‌​​‌​​​​​​‌‌​​​​‌​​‌​​​​​​‌‌​‌‌‌​​‌‌​​​​‌​‌‌​​‌​​​‌‌​‌​​‌​‌‌​​‌​‌​​‌​​​​​​‌‌​‌‌​​​‌‌​​‌​‌​​‌​​​​​​‌‌​​‌​​​‌‌​​‌​‌​‌‌​​​‌​​‌‌​‌‌‌‌