Customer Segmentation using K-Means Clustering (Mall Customers)

Every retailer knows customers are not all alike, but few can say precisely how their shoppers group — who spends freely, who buys carefully, who needs a nudge. Customer segmentation turns that intuition into data: group customers by behaviour so marketing, inventory and loyalty programs can target each group differently. The Mall Customers dataset is the classic teaching ground for this — 200 shoppers described by age, annual income and a spending score — and K-Means is the canonical first clustering algorithm. This project does it properly: standardized features, k-means++ initialization, the elbow method to justify k=5, silhouette-score validation, and five named, interpretable segments with business profiles. The interactive demo makes the abstract concrete — explore the segment map and find your own segment.

1. The 200-row Mall Customers dataset is loaded and inspected; Annual Income and Spending Score are selected as the clustering features.
2. Features are standardized (zero mean, unit variance) so income scale does not dominate the distance metric.
3. K-Means is run for k=1..10 with k-means++ initialization; the elbow in the WCSS curve justifies k=5.
4. The final model is fitted with k=5; the silhouette score is computed to quantify cluster separation.
5. Each cluster's centroid is translated into a business profile (Target, Careful, Spendthrift, Sensible, Standard) with a recommended marketing action.
6. In the web demo, the segment map renders all 200 customers coloured by cluster; the segment finder standardizes your inputs with the saved scaler and assigns the nearest centroid.

Tech stack:

• Python 3, scikit-learn (KMeans, metrics)
• Pandas, NumPy (data handling)
• Matplotlib (elbow plot, segment map)
• Jupyter Notebook (analysis & evaluation)
• HTML5 + JavaScript (interactive segment demo)
• Mall Customers dataset (200 records)

Dataset & model details

• Dataset: Mall Customers — 200 records with CustomerID, Gender, Age, Annual Income (k$/yr) and Spending Score (1–100). A standard public teaching dataset for clustering.
• Task: Unsupervised clustering; input = (annual income, spending score) standardized, output = segment assignment among 5 clusters.
• Model: K-Means with k=5 (elbow method), k-means++ initialization, 300 max iterations, features standardized with StandardScaler. Segment names and business profiles derived from centroid positions.
• Metrics: Silhouette score ≈ 0.55 (design target for the built-to-order run) and WCSS elbow analysis. Unsupervised learning has no accuracy — the project states this explicitly and validates with silhouette score plus business sense.

• [Elbow-method model selection] Within-cluster sum of squares plotted for k=1..10 with the elbow at k=5 — the standard, defensible way to choose the number of segments, not a guess.
• [Interactive segment map] All 200 customers plotted on income vs spending score, coloured by segment with centroids marked — hover-free, clean and presentation-ready.
• [Live segment finder] Enter age, annual income and spending score on sliders and get your segment instantly, with the distance to the centroid and a plain-English business profile.
• [Five interpretable segments] Target, Careful, Spendthrift, Sensible and Standard — each with a centroid profile and a marketing action (premium campaigns, value messaging, trend-driven offers...).
• [Silhouette validation] Silhouette score (≈ 0.55 design target) reported alongside the elbow analysis, so the clustering quality is quantified, not asserted.
• [Full analysis notebook] Data loading, scaling, K-Means fitting, elbow and silhouette analysis, segment profiling and visualization in one reproducible Jupyter notebook.
• [Segment profile cards] Per-segment income/spending centroids with business interpretation — ready to drop into the report and PPT.
• [Exported cluster model] Fitted scaler + K-Means centroids saved, so the demo assigns segments with the real model without refitting.

• Complete analysis Jupyter notebook (elbow, silhouette, profiling)
• Fitted scaler + K-Means model files
• Interactive segment-map web demo with live segment finder
• Elbow plot, silhouette analysis and segment-map figures
• Project report PDF (background, K-Means theory, methodology, segment profiles, business actions)
• PPT presentation for final review
• Viva Q&A preparation document (K-Means, elbow vs silhouette, standardization, unsupervised validation)

• Unsupervised learning has no ground-truth labels — segments are validated by silhouette score and business sense, never by accuracy.
• 200 records is a teaching-scale dataset; real retail segmentation uses far larger, richer customer data.
• Segment names (Target, Careful, ...) are interpretive labels on the clusters, not predictions about real people.
• The demo assigns segments with the fitted model; it does not retrain or discover new segments live.
• All metric figures are design targets for the analysis run, stated honestly — the report documents the actual achieved figures.