Customer churn is a formidable and persistent challenge for the telecommunications sector, as it significantly erodes profitability and impedes sustainable competitive advantage. Conventional machine learning methods often exhibit limitations in capturing non-linear intricacies and effectively managing voluminous, high-dimensional datasets. To address these issues, this investigation presents a novel hybrid predictive framework for enhanced customer churn prediction. Our proposed methodology integrates an optimized Multilayer Perceptron (MLP) with the Atomic Orbital Search (AOS) metaheuristic algorithm. Crucially, AOS is systematically deployed to fine-tune the critical connection weights and biases of the MLP architecture, mitigating prevalent issues like premature convergence and suboptimal parameter selection that are common in standalone neural networks. Empirical validation, conducted using the comprehensive Teldata dataset (7043 instances, 20 features) in MATLAB, unequivocally demonstrates the hybrid approach's superior efficacy over established conventional methodologies. Specifically, the hybrid MLP-AOS model achieved an impressive predictive accuracy of 97.8% on the test subset, a notable advancement compared to the 79% attained by Support Vector Machines (SVMs) and 75% by a conventional Multilayer Perceptron. These compelling findings underscore the proposed approach's capability in predicting customer churn with heightened precision, providing telecommunication management with an analytical instrument for identifying pivotal influencing factors and formulating effective retention strategies.