| dc.description.abstract | The focus of telecommunication companies has shifted from building a large customer base into
keeping customers in house. For these reasons, it is valuable to know which customers are likely to
switch to a competitor through porting out or purchasing a competitor line.
Since acqumng new customers IS more expensive than retaining existing customers, chum
prevention can be regarded as a popular way of reducing the company's costs. In this study, Cox
proportional hazard model and decision tree model are compared with conventional model.
The first model, the Cox model, is based on the theory of survival analysis, whereas the second
model, a decision tree, is commonly used in data mining. Both models are tested on a selection of
pre-paid customers from the database provided by Safaricom Limited.
Current conventional prediction used by Safaricom Limited was improved significantly by using
Cox proportional hazard and decision tree as they both performed better on the ROC curve.
However, for the duration under consideration decision tree performed better than Cox proportional
model.
Decision tree model selected gave probability of chum which is an improvement from conventional
model that only gives binary results of chum and not chum. Also, where the decision tree yields
approximately 50 percent probability of chum conventional model gave varying chum status | en |
