Data Transformation, Modelling and Prediction of Customer Churn using Deep Learning

Client prediction of churn is a difficult field of study that adds to client retention strategies. In a saturated and competitive market, telecommunications companies must rely on a viable churn customers prediction model to retain consumers. Previous research has focused on predicting customer attrition in the present moment or next month, when telecom businesses have not got enough time to create and implement churn control solutions. The predictive performance of present machine learning algorithms, which are frequently used by churn communities, appears to be at a stalemate, owing in part to the algorithms' weak feature extraction capacity. As a result, this research proposes a unique technique, a mix of neural networks with self- attention augmentation, to increase the efficiency of features filtering and feature extraction, hence enhancing model prediction performance. This study is being undertaken in the context of forecasting customer churn (CCP) in the telecommunications industry (TCI), wherein customer attrition is a typical occurrence. For the CCP problem, we developed a unique solution that combines data transformation approaches with machine learning models. We ran our tests on publicly accessible TCI datasets and evaluated performance using widely used assessment metrics (e.g., AUC, accuracy, recall, and F-measure). In this work, we used extensive comparisons to verify the impact of modification strategies.