Research Article
Evaluating Customer Churn Prediction with Machine Learning and Deep Learning Models
@INPROCEEDINGS{10.4108/eai.28-4-2025.2357829, author={Lavanya Peteti and Harsha Vardhini Kandivalasa and Jahnavi Sajja and Eva Patel}, title={Evaluating Customer Churn Prediction with Machine Learning and Deep Learning Models}, proceedings={Proceedings of the 4th International Conference on Information Technology, Civil Innovation, Science, and Management, ICITSM 2025, 28-29 April 2025, Tiruchengode, Tamil Nadu, India, Part I}, publisher={EAI}, proceedings_a={ICITSM PART I}, year={2025}, month={10}, keywords={customer churn prediction long short-term memory bidirectional long short-term memory support vector machine multi-layer perceptron}, doi={10.4108/eai.28-4-2025.2357829} }- Lavanya Peteti
Harsha Vardhini Kandivalasa
Jahnavi Sajja
Eva Patel
Year: 2025
Evaluating Customer Churn Prediction with Machine Learning and Deep Learning Models
ICITSM PART I
EAI
DOI: 10.4108/eai.28-4-2025.2357829
Abstract
Churn is one of the largest growth threats in the telecommunication industry. To manage it effectively, predictive modeling has become the strategy of choice. The following section of this paper compares some machine learning and deep learning techniques. The next part of this paper discusses a comparison of some machine learning and deep learning algorithms. In churn predictions the authors examined deep learning methods against other machine learning methods (Long Short-Term Memory - LSTM, Bi-directional LSTM - Bi-LSTM and Multi-Layer Perceptron - MLP, Support Vector Machine - SVM, Random Forest, Gradient Boosting, AdaBoost and Logistic Regression). The authors developed and evaluated the performances of deep learning and machine learning models with 667 instances in test set and 4568 instances in training set containing 20 features. The machine learning algorithms had the best accuracy from SVM and MLP, with AUC values of 0.8990 and 87.56% and 87.71% respectively. However, the deep learning methods surpassed the conventional algorithm accuracy with LSTM and Bi-LSTM having an accuracy of 91% and 90% respectively. Deep learning methods' great preprocessing, feature engineering and tuning capabilities allowed it to learn the behavior patterns of its customers better and show higher ability to manage churn and be a part of customer retention programs.
