METHODS FOR TRANSFORMING SEMISTRUCTURED DATA INTO RELATIONAL MODELS: CLASSIFICATION, APPLICATIONS, AND EVALUATION FOR ANALYTICS AND MACHINE LEARNING

The paper presents a systematic classification of methods for transforming semistructured data (XML and JSON) into relational models, focusing on their applicability for analytical tasks and machine learning. Four principal approaches are analyzed – structural, graphbased, semantic, and costbased. Each is assessed in terms of SQL query performance and training data preparation. Special attention is paid to semantic preservation, redundancy minimization, and achieving normalized data structures. The proposed classification supports the selection of optimal transformation techniques based on requirements for structural flexibility and computational efficiency.