Enhanced Robust Association Rules (ERAR)Method for Missing Values Imputation

Abstract

Missing values or incomplete data is a common problem that occurs in many applications. In most cases, recovering missing values from data sets is necessary to avoid bias conclusions made by omitting missing values. Missing values recovery (that is also known as missing values imputation) is an important research subject in the field of statistics and data mining. In this paper, we present the Enhanced Robust Association Rules (ERAR)method to extract useful association rules and avoid redundant rules. We show the enhancement made on ERAR to improve the imputation performed by the original Robust Association Rules (RAR). ERAR is designed in selecting the frequent items in datasets that are only related to missing values. Therefore, unnecessary frequent items can be ignored in generating the association rules. The result of the experiment shows that ERAR offers better performance in terms of the time taken for the imputation process and the amount of memory used to complete the imputation. In particular, ERAR behaves better in a monotone pattern of missing values than the arbitrary pattern. In terms of imputation accuracy, we found that both ERAR and RAR exhibit a decreasing rate of accuracy as the amount of missing values increases for data of arbitrary pattern, but this is not the case of data of the monotone pattern. With the findings, ERAR contributes to improving how one can deal with incomplete data.

Missing values imputation in Arabic datasets using enhanced robust association rules

“…In 1999 Ragel and Cremillex proposed missing values completion (MVC) using AR [12]. Based on MVC methods, several algorithms were developed, such as recycle combined association rules (RCAR) [13], fast recycle combined association rules (FRCAR) [14], association rule mining from data with missing values (ARDM) [15], Iterative missing-value completion [16], and enhanced robust association rules (ERAR) [17]. AR is a concept based on complete matching between values during the process of computing frequent itemsets.…”