Optimizing Churn Prediction Models: A Data Imbalance Handling Strategy for Enhanced Accuracy

In light of the accelerated expansion of the telecommunications sector and the intensifying competitive landscape, it would be advantageous to investigate potential strategies for reducing customer attrition. It is frequently the case that retaining customers represents a more cost-effective and profitable strategy than acquiring new ones. One potential method for reducing customer churn is to identify those customers who may be at risk through the utilization of customer churn prediction models. It may prove advantageous to consider the deployment of machine learning models, such as decision trees and random forests, for the purpose of prediction. However, one of the primary challenges in churn prediction is the presence of imbalanced data, wherein the number of customers who churn is significantly lower than those who remain. It would be advantageous to address the issue of unbalanced data in order to prevent the prediction model from misclassifying it. This research proposes a customer churn prediction model that addresses the issue of unbalanced data, encompassing the following stages: data understanding, data preparation, modeling, and evaluation. The findings indicate that addressing the issue of unbalanced data may lead to enhanced accuracy in the prediction model. For example, the accuracy of the decision tree increased from 75.3% to 79%, while that of the random forest increased from 78.8% to 83.3%. It is suggested that the proposed predictive model could help telecom companies to identify at-risk customers more effectively, thereby enabling them to take preventative measures to retain them. This could potentially result in improvements to customer retention and a reinforcement of competitiveness within the industry.