Missing Value Imputations by Rule-Based Incomplete Data Fuzzy Modeling

Abstract

Missing values are a common phenomenon in real-world datasets, which decreases the quality and reliability of data mining. Traditional regression-based imputation method estimates missing values through the relationship between attributes inferred by complete records. In order to describe the relationship more appropriately and make better use of present values, a rule-based incomplete data modeling method is proposed to impute missing values in this paper. The method utilizes incomplete records together with complete records for establishing Takagi-Sugeno (TS) models. In this process, the incomplete dataset is divided into several subsets and the linear functions containing only significant variables are built to describe the relationships between attributes in each subset. Experimental results demonstrate that the proposed method can effectively improve the performance of missing value imputation.