[Summary]
Sections | Key Points | |
Problem Definition | Importance of customer segmentation; the challenges in deploying proper models. | |
EDA | - Feature engineering (e.g., grouping ages, marital status, expenses).
- Dropping outliers for clearer insights.
- Positive correlation between income and expenses. | |
Proposed Model | - Elbow plot suggested K=3 clusters, but Gaussian Mixture Model (GMM) with K=5 was selected.
- Achieved detailed segmentation with an acceptable silhouette score of 0.1342. | |
Cluster Details | - Cluster 0: Low-income, low spenders; minimal engagement.
- Cluster 1: Moderate income and spending; family-oriented.
- Cluster 2: High-income, high spenders; highly engaged.
- Cluster 3: Moderate income; conservative spenders.
- Cluster 4: Diverse income and engagement; larger family sizes. | |
Business Solutions | - Cluster 0: Discount coupons, bulk purchase offers.
- Cluster 1: Buy-now-pay-later, family campaigns.
- Cluster 2: Premium offers, VIP loyalty programs.
- Cluster 3: Cashback events, essential bundles.
- Cluster 4: Household deals, segmented campaigns. |
Introduction:
•
Identifying distinct customer segments within a heterogeneous customer base was the primary challenge, which would enable the implementation of more targeted and effective marketing strategies.
Context:
•
A company requested a robust model to segment customers based on their behavioral and demographic characteristics, ensuring that marketing efforts could be precisely aligned with customer needs and preferences.
Exploratory Data Analysis
•
The established graphs and maps based on the current dataset did not provide the most meaningful insights as there were lots of overlaps
•
Therefore, feature engineering needed to be deployed
•
Also, the dataset underwent any possible outliers in age; then grouped the ages data into young, middle, and old groups.
•
Created new columns after grouping marital status, age, family size, expenses, total purchases, kids, engaged in days, and total accepted rate.
•
After dropping outliers, cleaning data, and doing feature engineering, further analysis was able to be made.
•
Positive correlation with income and expenses. High-income, high spending customers are the most valuable targets.
•
Single individuals possess higher average income. This may be because larger families have more financial responsibilities and burdens. Therefore, it will be important for the company to target single individuals as another main target.
•
The elbow plot was generated to determine optimal number of clusters. K = 3 seemed ideal but
After GMM decided with K =5
•
Did K-Means, hierarchical clustering, DBSCAN, PCA, and GMM
•
GMM model provided the best Segmentation with further granularity although with Slightly lower silhouette score of 0.1342 but acceptable
Insights:
Cluster 0
• Low-income, low spenders; minimal engagement, moderate family structures, minimal engagement
Cluster 1
• Moderate income, balanced spending and engagement, Moderate family structures
Cluster 2
• High-income, high spenders; highly engaged, individuals who tend to have fewer kids or teenagers at home
Cluster 3
• Moderate income, conservative spenders, moderate Income individuals
Cluster 4
• Diverse income, mixed spenders; varied engagement, Larger family sizes
Solutions:
Cluster 0:
• Diverse Discount Coupons
• Bulk and Group Purchase Offers:
• Free Shipping Day
Cluster 1:
• Buy Now, Pay Later Options:
• Family-Oriented Re-engagement Campaigns:
• Free Shipping Promotions:
Cluster 2:
• Premium and Exclusive Offers:
• VIP Loyalty Programs:
• Free Shipping on Premium Orders:
Cluster 3:
• Rational Cashback Events
• Essential Bundle Offers:
• Free Shipping Day:
Cluster 4:
• Larger Household and Special Deals:
• Segmented Campaigns:
• Free Shipping for Family Essentials