Predicting customer retention with ML models.
Customer retention is a critical part of an effective business model, whether it is for a streaming platform or clothing store. For companies, then, an important question is whether a customer will continue to purchase their services or not. Given data from a telecom company that has had some customers discontinue the service in the past months, this project aims to predict whether a customer will discontinue the service based on demographic geatures as well as company-related features such as the types of service they purchased.
In this project, a Random Forest Classifier was tested, chosen for its adeptness at handling categorical data and complex non-linear feature interactions, which are prevalent in the dataset. This modle also mitigates overfitting through its ensemble approach. To improve upon this, AdaBoost Classifier was used, which combines weak learners (decision stumps) for enhanced robustness and reduced hyperparameter tuning and reduced susceptibility to overfitting by weighing the learners based on their accuracy. Additionally, Neural Networks were explored to compare performance. Overall, Neueral Networks were found to be less effective than the Random Forest for the dataset probably due to its simplicity and size, and the AdaBoost Classifier was found to have the best result (highest ROC AUC score).