Churn prediction in Turkey's telecommunications sector: A proposed multiobjective–cost-sensitive ant colony optimization

Özmen M., Kızılkaya Aydogan E., Delice Y., Toksarı M. D.

Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, cilt.10, sa.1, 2020 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Derleme
  • Cilt numarası: 10 Sayı: 1
  • Basım Tarihi: 2020
  • Doi Numarası: 10.1002/widm.1338
  • Dergi Adı: Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, Computer & Applied Sciences, INSPEC, Library, Information Science & Technology Abstracts (LISTA)
  • Anahtar Kelimeler: churn prediction, cost-sensitive learning, feature selection, high dimension, CLASS IMBALANCE PROBLEM, MACHINE LEARNING TECHNIQUES, HANDLING CLASS IMBALANCE, CUSTOMER CHURN, FEATURE-SELECTION, GENETIC ALGORITHM, OVERSAMPLING TECHNIQUES, NSGA-II, CLASSIFICATION, INDUSTRY
  • Erciyes Üniversitesi Adresli: Evet

Özet

© 2019 Wiley Periodicals, Inc.Players in the telecommunications sector struggle against the competition to keep customers, and therefore they need effective churn management. Most classification algorithms either ignore misclassification cost or assume that the costs of all incorrect classification errors are equal. But as in real life, many classification problems have different misclassification costs and this difference cannot be ignored. For this reason, studies on cost-sensitive classification approaches have gained importance in recent years. The characteristics of telecommunications datasets such as high dimensionality and imbalance are making it difficult to achieve the desired performance for churn prediction. By taking this into consideration, we propose a multiobjective–cost-sensitive ant colony optimization (MOC-ACO-Miner) approach which integrates the cost-based nondominated sorted genetic algorithm feature selection and multiobjective ACO based cost-sensitive learning. MOC-ACO-Miner is applied to one of Turkey's top 100 information technology companies for customer churn-prediction. Finally, experiments find out that the model performs quite well with the area under receiver operating characteristic curve values of 0.9998 for predicting churners and therefore it can be beneficial for the highly competitive telecommunications sector. This article is categorized under: Algorithmic Development > Association Rules Application Areas > Industry Specific Applications Technologies > Prediction Technologies > Data Preprocessing.