Deciding the Best Machine Learning Algorithm for Customer Attrition Prediction for a Telecommunication Company

摘要

Customers are so important in business that every firm should put great effort into retaining them. To achieve that with some measure of success, the firm needs to be able to predict the behaviour of their customers with respect to churn or attrition. There are many machine learning algorithms that may be used to predict attrition, but this paper considers only four of them. Logistic regression, k-nearest neighbour, random forest and XGBoost machine learning algorithms were applied in different ways to the dataset gotten from Kaggle in order to decide the best algorithm to suggest to the company for customer attrition prediction. Results showed that the logistic regression or random forest algorithm may be adopted by the telecom company to predict which of their customers may leave in the future based on their recall and precision scores as well as the AUC values of approximately 75%. The logistic regression algorithm gave metrics of 73% accuracy and 59% f1-score while the random forest algorithm yielded 70% accuracy and a 58% f1-score. However, it was also suggested that if the choice of model was based on accuracy and f1-score, the logistic regression model would be the best to be adopted.