The increasing complexity and pervasive legacy infrastructure in enterprise computing have catalyzed the exploration of innovative approaches to system modernization. Traditional methods of refactoring and modularization have often proven insufficient due to scale, heterogeneity, and incomplete system documentation. Recent research emphasizes leveraging machine learning techniques to enhance service boundary detection, enabling more effective migration from monolithic architectures to modular or service-oriented designs. This study synthesizes existing theoretical frameworks on object-oriented design metrics, coupling, cohesion, and maintainability with contemporary machine learning-assisted approaches to system decomposition. Specifically, the paper examines the application of predictive analytics for identifying modular boundaries within legacy systems, alongside strategies for integrating service-oriented principles and microservices patterns. Through a comprehensive review of literature and interpretive analysis, we highlight the implications of automated boundary detection on system maintainability, resilience, and adaptability. We also discuss methodological considerations, including data collection from change management repositories, metric validation, and the limitations of current models. By critically examining the convergence of software metrics, empirical studies, and machine learning models, this research contributes to a nuanced understanding of how high-volume legacy systems can evolve toward more flexible, maintainable, and strategically aligned architectures. The study concludes with recommendations for integrating machine learning into legacy system modernization workflows, and outlines directions for future research in predictive modularization, automated refactoring, and enterprise-scale deployment of modular architectures.