Comparative Study of Imputation Algorithms Applied to the Prediction of Student Performance

Abstract

Student performance and its evaluation remain a serious challenge for education systems. Frequently, the recording and processing of students’ scores in a specific curriculum have several flaws for various reasons. In this context, the absence of data from some of the student scores undermines the efficiency of any future analysis carried out in order to reach conclusions. When this is the case, missing data imputation algorithms are needed. These algorithms are capable of substituting, with a high level of accuracy, the missing data for predicted values. This research presents the hybridization of an algorithm previously proposed by the authors called adaptive assignation algorithm (AAA), with a well-known technique called multivariate imputation by chained equations (MICE). The results show how the suggested methodology outperforms both algorithms.