Churn prediction model using variable churn window for online purchases
The identification of retainable online non-contractual customers is pertinent for the operations and growth of non-contractual online businesses, since there are no obligations for customers to be loyal to a particular online business. This research, therefore, was aimed at proposing a generic chur...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2022
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The identification of retainable online non-contractual customers is pertinent for the operations and growth of non-contractual online businesses, since there are no obligations for customers to be loyal to a particular online business. This research, therefore, was aimed at proposing a generic churn prediction model that online businesses may utilise. Firstly, the attributes that can be used to predict customer churn for non-contractual customers were identified. The traditional attributes being Recency(R), Frequency (F) and Monetary, (M) were further expanded to include another attribute to improve the prediction model. This attribute, known as “Periodicity” was introduced as a unique attribute for each customer using the historical buying pattern of the customer within a certain period. Secondly this research explored the use of 2 models, namely Easy Prediction Model (EPM) and Automated Prediction Model (APM) for customer churn prediction in a noncontractual setting. EPM is a simple churn prediction model whereas APM uses Supervised Machine Learning techniques to make a customer churn prediction. This research also proposed a new churn window called Variable Churn Window (VCW), which was derived based on individual customers’ purchase history trends. VCW is different from the traditional definition of churn window, whereby a defined churn window is applied across customers regardless of their individual purchasing history. Generally, customers have their own unique churn window based on their purchasing behaviour. Both EPM and APM use the VCW in their respective churn predictions model. Thirdly this research also explored how these models will affect the prediction of customer churn behaviour. Results revealed that the proposed churn window model has better accuracy compared with the traditionally defined window which is called n-th Months Churn Window (nMCW). This leads to the conclusion that the proposed prediction models which utilises the new churn window, VCW, can be useful to support marketing strategies and activities of non-contractual online businesses. As for the APM, by adding the attribute “Periodicity to the traditional RFM attributes and using the combination of VCW, the prediction model has improved. This is proven by using the Area Under the Curve (AUC) as the evaluation metrics. Furthermore, it can be concluded that APM using the Random Forest has the best AUC values based on the Superstore dataset. |
|---|