Customer retention model using machine learning for improved user-centric quality of experience through personalised quality of service

Customer retention remains a critical challenge in subscription-based industries, particularly in telecommunications, where evolving customer expectations and traditional churn prediction models relying on static behavioural data fail to capture the dynamic nature of customer sentiment. This study addresses these limitations by developing a Personalised Quality of Service (PQoS) framework, integrating advanced natural language processing (NLP) models-RoBERTa and BERT- with machine learning algorithms to analyse customer feedback and predict churn more effectively. The study utilises a novel approach that combines sentiment analysis, contextual embeddings, and predictive modelling to enhance churn prediction accuracy. By incorporating Expected Utility Theory (EUT), the framework evaluates and prioritises retention strategies based on their expected impact on customer satisfaction and loyalty. The empirical analysis, conducted using five machine learning models, demonstrates that the Neural Network model with RoBERTa outperforms traditional methods, achieving 96% accuracy, 97% F1-score, 88% precision, and 95% recall. This research contributes to the customer relationship management (CRM) Framework by re-conceptualising churn prediction as a sentiment-driven process and introducing EUT-based decision-making in customer retention. The findings offer businesses a scalable, data-driven strategy for enhancing customer engagement, reducing churn, and sustaining long-term value creation.