Data Mining Approach for Predicting Learner's Achievement

Abstract

Student achievement variables that may be included into student database can be classified into three main categories, student variables. Instructor variables and general variables. This paper presents a new machine-learning model for extracting knowledge From student attributes in a given database. This knowledge can be used for determining the relative importance and effectiveness of student's attributes for the prediction of their college academic achievement, and the relationship between these attributes and their achievement. The model includes three main algorithms namely: preprocessing of database, attribute selection and rule extraction algorithm. Preprocessing of database aims to alleviate the dimensionality of the given database. It is performed according to (i) Detecting memo attributes and abstracting their field values into minimum abstraction level, (ii) Detecting the attributes, which have repeated values (including sparse values), and dropping them from database and (iii) Using fuzzification for transferring the attributes of continuous values into linguistic terms. This transformation leads to reducing the search space. Attribute selection algorithm selects the most relevant attributes set by the calculations of an evaluation function. The resulted set of attributes is passed to rule extraction algorithm for extracting an accurate and comprehensible set of rules.