Importance of Customer Churn Prediction Using Machine Learning and Providing Recommendation Plans in Mobile Telecommunication Industry in Sri Lanka

Abstract

This research study is focusing on predicting customer churn using machine learning in the Sri Lanka telecommunication industry. This model can apply to the telecommunication sector of Sri Lanka to manage or early identification of the customer churn which has major commercial impact in this industry. The review of the literature is referred to identify the previous related to customer churn and predictive model development and how research works were conducted to predict customer churn prediction using machine learning and other statistical methodologies. And what significant factors contribute and what are the algorithms used and performance and accuracy of the output were considered from published academia. This research is done using Cell2cell openly available dataset in the telecommunication domain which is a standard featured dataset which can be generalized to the Sri Lanka Telecommunication industry based on the assumption of generalization. Dataset is subjected to preprocessing with missing value and outlier treatments. Filter method is used for the feature selection using Chi-square and Anova for numerical and categorical variables. The research study will be conducted using supervised learning algorithms Naïve Bayes (NB), Decision Trees (DT), and Support Vector Machines (SVM). Here single classifiers performance is analyzed. After that under the voting method ensembled classifiers are formed using two singles classifiers NB &DT, DT & SVM, SVM & DT and performance are evaluated. Performance of the classifiers are examined using ROC curve and Discrete statical method in this research. Overall accuracy of the predictive models is the key output parameter which is related to customer churn which was analyzed by Discrete statical. All the features of the data set visually represent with histogram and box plot tools to get an overview of the features. Application developed on distributed architecture which data set and predictive model is running on server and remotely admin and user interface can connect to it. Admin interface is supported for deep analysis of dataset and user interface can be used for future prediction.