This study aimed to transform customer segmentation in the retail sector by systematically evaluating and contrasting various clustering algorithms to generate deeper market insights. The preprocessing phase integrates key data preparation techniques, including mean imputation to address missing values, Z-Score Standardization for normalization, and One-Hot Encoding to manage categorical data. To counter data imbalance, the Adaptive Synthetic Sampling (ADASYN) technique was employed. Dimensionality reduction is optimized through enhanced Principal Component Analysis (PCA), ensuring computational efficiency while retaining critical data attributes. Feature extraction encompasses a comprehensive range of metrics, including statistical measures, higher-order statistical descriptors, entropy features, and correlation-based parameters, thereby ensuring a holistic representation of customer data. At the core of this research is the advanced Evofusion model, an innovative ensemble deep learning framework. Evofusion synergizes the strengths of three neural architectures, EfficientNet, SqueezeNet, and MobileNetV2, to enhance the accuracy and robustness of customer segmentation. By leveraging the complementary capabilities of these models, Evofusion offers a sophisticated solution that surpasses the traditional methods in uncovering actionable insights. This novel approach redefines retail customer analytics by integrating cutting-edge techniques into data preprocessing, feature engineering, and ensemble modeling. This research not only advances segmentation methodologies but also provides a scalable and adaptable framework for businesses to understand and target their customer base with unprecedented precision. These results underscore the potential of deep learning ensembles to revolutionize data-driven decision-making in the retail industry..