Airline Customer Segmentation Based on LRFMC Model

Introduction

This is an unsupervised machine learning practice on customer segmentation.

Objective

To identify the customer segmentation in aviation company, analyze characteristics of different clusters and suggest business strategy to customer categories of different values.

Data Understanding

This dataset has 62988 rows dan 23 columns. Column names themselves are self explanatory, however here I provide several columns explanation related to aviation industry:

1. MEMBER_NO: ID Member
2. FFP_DATE: Frequent Flyer Program Join Date
3. FFP_TIER: Frequent Flyer Program Tier
4. LOAD_TIME: Observed Time
5. FLIGHT_COUNT: Customer Flight Count
6. BP_SUM: Itinerary
7. SUM_YR_1: Fare Revenue
8. SUM_YR_2: Votes Prices
9. SEG_KM_SUM: Total flight kilometers in observation window
10. LAST_FLIGHT_DATE: Last Flight Date
11. LAST_TO_END: The last flight time to the end of the observation window
12. AVG_INTERVAL: Average flight interval
13. MAX_INTERVAL: Maximum flight interval
14. EXCHANGE_COUNT**: Exchange count
15. avg_discount: Average discount rate
16. Points_Sum: The number of points earned by the customer
17. Points_NotFlight: Point not used by customers

Method

I used LRFMC Model that is commonly used in aviation dataset. LRMFC stands for Length, Recency, Frequency, Monetary and Discount Factor. The details of each parameter is elaborated below:

1. Length: the length of days between subscriber's first register day to observation time. A larger number means they have been a member for a longer time
2. Recency: the length of a passenger's last consumption to observation window. A smaller number means they flew more recently
3. Frequency: the passenger's consumption frequency within a certain period of time. A bigger number means they flew more frequently
4. Monetary the average amount spent over a certain period of time. A bigger number means they spent more money
5. Discount Factor: the average space discount factor for passengers traveling within a certain period of time A bigger number means they use discounts more After looking at the data, we use the columns below that related to the LRFMC features:

• Length: LOAD_TIME - FFP_DATE
• Recency: LAST_TO_END
• Frequency: FLIGHT_COUNT
• Monetary: SEG_KM_FM
• Customer Cost: avg_discount

K-means clustering method is used to classify the customer segmentation.

Conclusion

From elbow plot, there is a possibility that the optimum clusters are 4-6. After consider the business perspective, we decide that the suitable number of clusters are 5. The clusters are:

1. Cluster 1 - The Hibernating Customers
2. Cluster 2 - The Promising Customers
3. Cluster 3 - The Loyal Customers
4. Cluster 4 - The Old-Seasonal Customers
5. Cluster 5 - The Cheap Customer

To see details on what are the characteristics of each cluster and recommendation for the company, make sure to check the notebook and power point files above.