Dynamic customer churn prediction strategy for business intelligence using text analytics with evolutionary optimization algorithms

•Propose a new Dynamic Customer Churn Prediction model for Business Intelligence.•Apply Text Analytics with Metaheuristic Optimization algorithm for Churn Prediction.•Design a new Chaotic Pigeon Inspired Optimization based Feature Selection technique.•Employ Parameter Tuned LSTM-SAE model to classify the feature reduced data.•Validate the classification performance on benchmark churn prediction dataset. In the digital era, innovations in business intelligence are critical to staying competitive and popular across the growing business trends. Businesses have begun to investigate the next stage of data analytics and business intelligence solutions. On the other hand, Customer Churn Prediction (CCP) is a crucial process in business decision making, which properly identifies the churn users and takes necessary steps for customer retention. churn and non-churn customers have resembling features. Therefore, this research work designs a dynamic CCP strategy for business intelligence using text analytics with metaheuristic optimization (CCPBI-TAMO) algorithm. In addition, the chaotic pigeon inspired optimization based feature selection (CPIO-FS) technique is employed for the feature selection process and reduces computation complexity. Besides, long short-term memory (LSTM) with stacked auto encoder (SAE) model is applied to classify the feature reduced data. In the LSTM-SAE model, the ability of SAE in the detection of compact features is integrated into the classification capability of the LSTM model. Finally, the sunflower optimization (SFO) hyperparameter tuning process takes place to further improve the CCP performance. A detailed simulation analysis is performed on the benchmark customer churn prediction dataset and the experimental values highlighted the superior performance of the proposed model over the other compared methods with the maximum accuracy of 95.56%, 93.44%, and 92.74% on the applied dataset 1-3 respectively.