FIONA: Detecting Outliers in Attributes of Relational Datasets with Categorical Values

Outlier detection plays a vital role in data cleaning, significantly impacting analysis and decision-making processes. Although extensive research has been conducted on numerical outlier detection, identifying outliers in relational data with categorical attributes presents unique challenges due to the complexity of defining an appropriate similarity measure. Existing methods often rely on converting categorical values to numerical ones, using frequency as an indicator of outlierness, and extracting predefined syntactic structures from the values.In this paper, we introduce FIONA (FInding Outliers iN Attributes), a novel approach designed to detect outliers in relational data with categorical values in distributed scenarios. Usually, categorical values in relational datasets exhibit specific syntactic structures. FIONA establishes a similarity measure to uncover hidden patterns and identify dominant patterns within the data. Values that deviate from these dominant patterns are reported as outliers.Compared to existing tools, FIONA excels in accurately identifying outliers and dominant patterns in distributed datasets and offers clear and concise explanations for each identified outlier. Our approach ensures that distributed systems can maintain data integrity and reliability, enhancing the effectiveness of datadriven decision-making in large-scale, distributed environments.