Enhanced Churn Prediction in Telecom using PCA-based Feature Reduction and Bagging Ensembles

Ongoing technological advancements have produced efforts for a competitive scenario among laboratories pertaining to the technical domain. In such instances, the telecommunications sector is no exception. There is a huge competition among telecommunications service providers to achieve maximum profits. Hence, customer migration predictions are a major concern for the telecom sector and influence customer loyalty plans and profitability. Wave prevention becomes a strategic priority as telecommunication companies are confronted with challenges like rising competition, evolving customer expectations, and new customer acquisition costs. This research formulated a predictive model that helps us to determine the customers who are most likely to shift the service provider or churn the company in near future. A major hurdle in customer churn prediction lies in handling the imbalance or skewed data within the dataset that employs ensemble methods to rectify the issue. The machine learning models used for training the data in this research paper are Random Forest Classifier, Decision Tree Classifier, SVM, and LightBGM. PCA was used to reduce dimensions, whereas bagging methods were implemented to enhance the robustness of the model. Experimental findings exhibit predictive models that identify potential paths with optimal accuracy.