<![CDATA[Relational databases form the backbone of modern information systems. However, data quality issues, such as duplicate records, invalid formats, missing values, and cross-table inconsistencies, can significantly reduce the accuracy of data-driven decision-making. Conventional rule-based validation is effective for detecting structured errors but has limited capability in identifying ambiguous errors, such as typographical variations in entity names. This study proposes a hybrid rule-based system that combines SQL triggers for structured error detection with fuzzy matching using the RapidFuzz Python library to identify semantically similar records across multiple relational database tables. The proposed system was implemented using six primary database tables, six corresponding quarantine tables, and a centralized error_log table. The evaluation was conducted using a synthetic dataset containing 2,775 records distributed across the six tables. The dataset was systematically generated using the generate_dataset.py script, with various intentionally injected data quality issues to enable accurate verification of the detection results. The experimental results show that the proposed system detected 472 data quality issues, with 297 records automatically moved to the quarantine tables. The rule-based component identified 311 errors (65.9%), including format violations, negative values, and referential integrity violations. Meanwhile, the fuzzy matching component detected 127 semantic errors that could not be identified using SQL rules alone, including 112 duplicate customer names, 7 similar product names, and 5 inconsistent product categories. On the experimental dataset, the proposed hybrid approach detected 34.1% more data quality issues than a rule-based validation approach alone. These findings demonstrate that integrating rule-based validation with fuzzy matching substantially improves error detection capability in relational databases, particularly for semantic inconsistencies that are difficult to capture using conventional validation rules.]]>
Copyrights © 2026
