Domain knowledge integration in data mining for churn and customer lifetime value modelling: new approaches and applications

The evaluation of the relationship with the customer and related benefits has become a<br/>key point for a company’s competitive advantage. Consequently, interest in key<br/>concepts, such as customer lifetime value and churn has increased over the years.<br/>However, the complexity of building, interpreting and applying customer lifetime value<br/>and churn models, creates obstacles for their implementation by companies. A proposed<br/>qualitative study demonstrates how companies implement and evaluate the importance<br/>of these key concepts, including the use of data mining and domain knowledge,<br/>emphasising and justifying the need of more interpretable and acceptable models.<br/>Supporting the idea of generating acceptable models, one of the main contributions of<br/>this research is to show how domain knowledge can be integrated as part of the data<br/>mining process when predicting churn and customer lifetime value. This is done<br/>through, firstly, the evaluation of signs in regression models and secondly, the analysis<br/>of rules’ monotonicity in decision tables. Decision tables are used for contrasting<br/>extracted knowledge, in this case from a decision tree model. An algorithm is presented,<br/>which allows verification of whether the knowledge contained in a decision table is in<br/>accordance with domain knowledge. In the case of churn, both approaches are applied<br/>to two telecom data sets, in order to empirically demonstrate how domain knowledge<br/>can facilitate the interpretability of results. In the case of customer lifetime value, both<br/>approaches are applied to a catalogue company data set, also demonstrating the<br/>interpretability of results provided by the domain knowledge application. Finally, a<br/>backtesting framework is proposed for churn evaluation, enabling the validation and<br/>monitoring process for the generated churn models.

Name dc:type.qualificationname

Ph.D.

Level dc:type.qualificationlevel

doctoral

Grantor dc:publisher.institution

University of Southampton

Year dc:date.issued

2009

Author dc:creator

• de Oliveira Lima, Elen

Advisors dc:contributor.advisor

• Baesens, Bart
• Mues, Christophe