Customer Segmentation Using K-Means Clustering

Author: Zaki Abiyu Aqilah
Tools: R (ggplot2, kmeans, factoextra)
Goal: Segment customers into distinct groups based on demographic and spending behavior to enable targeted marketing strategies

---

Project Overview

This project applies unsupervised machine learning (K-Means clustering) to segment customers into five distinct groups using their demographic profile (gender, age, profession, residence type) and annual spending. The model helps businesses tailor marketing campaigns, optimize product offerings, and improve customer retention.

Key Results:

• Between-SS / Total-SS: 92.4% (excellent cluster separation)
• 5 distinct customer segments identified
• Highest spending segment: Cluster 5 (Silver Mid Professional) — ~8.8 million Rupiah/year

---

Dataset

The dataset contains 50 customer records with the following variables:

Variable	Type	Description
Customer_ID	Character	Unique customer identifier
Nama Pelanggan	Character	Customer name
Jenis Kelamin	Categorical	Gender (Pria/Wanita) → numeric mapping (1 = Pria, 2 = Wanita)
Umur	Numeric	Age in years
Profesi	Categorical	Profession (mapped to integers 1‑5)
Tipe Residen	Categorical	Residence type (Cluster/Sector) → numeric (1 = Cluster, 2 = Sector)
NilaiBelanjaSetahun	Numeric	Annual spending (in Rupiah, converted to millions)

The original Text file is not included in this repository. See data/README.md for data structure details.

---

Clustering Result

Optimal Number of Clusters (Elbow Method)

The Elbow Method was applied to determine the optimal k for K-Means clustering. The total within-cluster sum of squares (WSS) was calculated for k = 1 to 10. The elbow is clearly observed at k = 5, where the reduction in WSS starts to diminish.

k (clusters)	Total Within-Cluster SS	Interpretation
1	2,551.74	-
2	1,329.98	Sharp drop
3	995.83	Continued drop
4	810.71	Continued drop
5	682.89	Elbow point → optimal k
6	608.81	Marginal improvement
7	539.85	Marginal improvement
8	469.26	Marginal improvement
9	435.77	Marginal improvement
10	393.60	Marginal improvement

Between_SS / Total_SS = 92.4% – This indicates that the 5-cluster solution explains 92.4% of the total variance, a very strong separation.

Final Clusters (k=5)

Cluster	Size	Gender1	Avg Age	Profession2	Residence3	Avg Spending (Mio Rupiah)	Segment Name
1	12	1.75 (Mixed)	31.58	3.92 (Professional mix)	1.25 (Mostly Cluster)	7.33	Silver Youth Gals
2	10	1.70 (Mixed)	52.50	3.80 (Professional mix)	1.30 (Mostly Cluster)	6.02	Diamond Senior Member
3	5	1.40 (Mostly Male)	61.80	4.20 (Entrepreneurial)	1.40 (Mixed)	8.70	Gold Young Professional
4	14	2.00 (All Female)	20.07	3.57 (Student mix)	1.36 (Mixed)	5.90	Diamond Professional
5	9	2.00 (All Female)	42.33	4.00 (Professional mix)	1.56 (Mixed)	8.80	Silver Mid Professional

¹ Gender mapping: 1 = Pria (Male), 2 = Wanita (Female)
² Profession mapping: 1 = Ibu Rumah Tangga, 2 = Mahasiswa, 3 = Pelajar, 4 = Professional, 5 = Wiraswasta
³ Residence mapping: 1 = Cluster, 2 = Sector

Segment Profiles & Business Insights

Cluster 1 – Silver Youth Gals

• Profile: Young adults (avg 32 years), mixed gender, professionals/entrepreneurs, moderate spending (~7.3M).
• Recommendation: Target with digital campaigns, lifestyle products, and social media engagement.

Cluster 2 – Diamond Senior Member

• Profile: Seniors (avg 53 years), mixed gender, stable spending (~6.0M).
• Recommendation: Offer loyalty programs, healthcare products, and retirement planning services.

Cluster 3 – Gold Young Professional

• Profile: Mostly male seniors (avg 62 years), entrepreneurs, high spending (~8.7M).
• Recommendation: Premium/exclusive products, personalized services, partnership opportunities.

Cluster 4 – Diamond Professional

• Profile: Young females (avg 20 years), students/early professionals, moderate spending (~5.9M).
• Recommendation: Build long-term loyalty via educational content, discounts, and community events.

Cluster 5 – Silver Mid Professional

• Profile: Middle-aged females (avg 42 years), professionals/entrepreneurs, highest spending (~8.8M).
• Recommendation: Prioritize VIP treatment, exclusive offers, and upselling campaigns.

Download detailed outputs: Cluster centers (CSV) | Segment summary (CSV) | Segment summary (TXT)

---

Predictions New Customers

The trained K-Means model can predict the segment for new customers using distance to cluster centers. Below are sample predictions:

Application	Gender	Age	Profession	Residence	Spending (Million Rupiah)	Predicted Segment
App 1	Female	20	Pelajar	Cluster	3.5	Diamond Professional
App 2	Male	45	Wiraswasta	Sector	9.0	Gold Young Professional
App 3	Female	35	Professional	Cluster	7.0	Silver Youth Gals

---

How to Run the Code

1. Clone this repository
2. Place your data file (CustomerSegmen.txt) in the data/ folder
3. Run script/model.R in RStudio or R console

---

Connect with Me