Detection of outlier information using linguistically quantified statements – the state of the art

Abstract

Over the past decade, outlier detection has become a prominent area of research. Much of the current literature on Knowledge Discovery in Databases (KDD) pays particular attention to the identification of rare, unusual data items or surprising observations that differ substantially from the majority of the data. In Data Mining (DM), outlier detection refers to the identification of unusual data records that might be interesting in the context of data pattern discovery or may represent potential data errors that require further investigation. Recent research has emphasized the need for KDD / DM systems capable of processing information given in a linguistic form and communicating the result to the user in a natural language. Such systems represent a valuable tool for intelligent data analysis tasks, including outlier detection. The paper provides an overview of the current state of research on the detection of outlier information using linguistically quantified statements based on diverse variants of fuzzy linguistic summarization. It pays special attention to the use of monotonic and non-monotonic quantifiers defined on interval-valued fuzzy sets. The outlier detection procedure using linguistic summaries is described in detail.