Model-Driven Engineering and Machine Learning for Legacy System Modernization

Model-driven engineering is significant in modern software development, as it leverages models to guide software design and implementation. However, legacy systems are characterized by outdated architectures and poor documentation, which pose challenges for maintenance. Model-Driven Reverse Engineering (MDRE) addresses these issues by extracting models from legacy systems, enabling better comprehension. This study explores clustering techniques to extract high-level concepts from source code. Clustering reveals system concepts that can be transformed into models by grouping similar code entities. These models provide a foundation for system modernization and further development. Unlike existing MDRE approaches, this study systematically evaluates multiple clustering techniques and preprocessing scenarios to assess their ability to recover high-level system concepts from legacy source code. The results highlight the impact of preprocessing and noise on concept extraction, providing insights for applying clustering within model-driven modernization workflows.