An Appraisal Model Based on a Synthetic Feature Selection Approach for Students’ Academic Achievement

Abstract

Obtaining necessary information (and even extracting hidden messages) from existing big data, and then transforming them into knowledge, is an important skill. Data mining technology has received increased attention in various fields in recent years because it can be used to find historical patterns and employ machine learning to aid in decision-making. When we find unexpected rules or patterns from the data, they are likely to be of high value. This paper proposes a synthetic feature selection approach (SFSA), which is combined with a support vector machine (SVM) to extract patterns and find the key features that influence students’ academic achievement. For verifying the proposed model, two databases, namely, “Student Profile” and “Tutorship Record”, were collected from an elementary school in Taiwan, and were concatenated into an integrated dataset based on students’ names as a research dataset. The results indicate the following: (1) the accuracy of the proposed feature selection approach is better than that of the Minimum-Redundancy-Maximum-Relevance (mRMR) approach; (2) the proposed model is better than the listing methods when the six least influential features have been deleted; and (3) the proposed model can enhance the accuracy and facilitate the interpretation of the pattern from a hybrid-type dataset of students’ academic achievement.