Abstract
In this paper we present a new approach to predicting telecommunications churn. Churn prediction can be considered as a multi-objective optimization problem, where the accuracy of predicting both churning and staying consumers need to be optimized simultaneously. As the existing classification methods failed to produce balanced solutions, we developed a new multi-population genetic algorithm for the induction of decision trees. By introducing multiple populations, linear ranking selection and adequate fitness function we were able to avoid overly biased solutions. The evaluation results of our algorithm’s performance in comparison with the existing methods show that it was able to find highly accurate and balanced solutions. DOI: http://dx.doi.org/10.5755/j01.eee.19.6.4578
Highlights
Telecommunications churn – moving to a different telecommunications company – today is still a major deal within companies
We propose a multipopulation genetic algorithms (MPGA) for the induction of decision tree (DT) that includes two subpopulations (Fig. 1)
In this paper we presented a new multi-population genetic algorithm (MPGA) for the induction of decision trees, which has been applied and thoroughly evaluated on the problem of predicting telecommunications churn
Summary
Telecommunications churn – moving to a different telecommunications company – today is still a major deal within companies. Considering the telecommunications churn problem, the aim of classification is to learn to predict whether a consumer will move to a different company based on the consumers’ data stored within company’s database. In this study we have designed and evaluated a special kind of GA for DT induction, namely a multi-population cross-selection GA, that was aimed at optimizing DTs with regard to overall prediction accuracy simultaneously with preserving the balance between positive (customers who churned) and negative cases (customers who stayed with a company). In the case of telecommunications churn problem set they were not able to produce the expected results and were not much different than the rest of the classification methods used in our experiment – they all optimized overall accuracy at the expense of low classification accuracy of positive cases (customers who churned), which were a real minority among all cases (less than 15%)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
