A Relative Tolerance Relation of Rough Set in Incomplete Information

Abstract

University is an educational institution that has objectives to increase student retention and also to make sure students graduate on time. Student learning performance can be predicted using data mining techniques e.g. the application of finding essential association rules on student learning base on demographic data by the university in order to achieve these objectives. However, the complete data i.e. the dataset without missing values to generate interesting rules for the detection system, is the key requirement for any mining technique. Furthermore, it is problematic to capture complete information from the nature of student data, due to high computational time to scan the datasets. To overcome these problems, this paper introduces a relative tolerance relation of rough set (RTRS). The novelty of RTRS is that, unlike previous rough set approaches that use tolerance relation, non-symmetric similarity relation, and limited tolerance relation, it is based on limited tolerance relation by taking account into consideration the relatively precision between two objects and therefore this is the first work that uses relatively precision. Moreover, this paper presents the mathematical properties of the RTRS approach and compares the performance and the existing approaches by using real-world student dataset for classifying university’s student performance. The results show that the proposed approach outperformed the existing approaches in terms of computational time and accuracy.