Churn Rate Prediction in Telecommunications Companies

• Autor(a) principal: Lima, Ana Lúcia de Morais
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10400.6/13269
• topic: Churn Rate; Data Science; Machine Learning; Predictive Model; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Customer churn is a central concern for companies operating in industries with low switching costs. Among all industries, the one that suffers most from this problem is the telecommunications sector, with an annual churn rate of approximately 30%. As operators grow, so does the volume of data, and understanding and interpreting this data is necessary for operators to understand why customer churn is happening. Through data science, machine learning, and artificial intelligence techniques, the possibilities of predicting customer churn have increased significantly. In this research, the proposed methodology consists of six phases. In its first phases, data preprocessing and feature analysis are performed. In the third phase, feature selection is performed. Then, the data were divided into two parts of training and testing, in the proportion of 80% and 20%, respectively. For the prediction process, the most popular prediction models were applied, i.e. logistic regression, vector machine, naive bays, random forest, decision trees, etc. In the training set, boosting and ensemble techniques were applied to achieve better model accuracy. In the training set, K­fold cross­validation was used to avoid overlapping models. The results are evaluated using the confusion matrix and the AUC curve. The Adaboost, Catboost and XGBoost classifiers obtained the highest accuracy in the range of 85% and 92%. The highest AUC score was 98% obtained by Random Forest and 93% XGBoost which outperformed the other models.