Adapting The Score Prediction to Characteristics of Undergraduate Student Data

Thong Le Mai, Phat Thanh Do, Minh Thanh Chung, Van Thanh Le, Nam Thoai

Abstract: Educational Data Mining (EDM) is necessary for extracting useful information from educational institutions. Various algorithms have been employed to measure student's GPA in the next semester's courses. The results can be used to early identify dropout students or help students choose the elective courses which are appropriate for them. Some of the most widely used methods for predicting students performance in their future courses are based on techniques frequently used in Recommender System such as Collaborative Filtering and Matrix Factorization. Consequently, the prediction method can affect the accuracy of the prediction results. However, the characteristics of the dataset also have a large impact on the performance of the corresponding prediction model. In this paper, we will analyze Ho Chi Minh University of Technology (HCMUT) dataset which consists of information about undergraduate students from 2006 to 2017 by conducting various experiments on two distinct faculties using multiple types of collaborative filtering, matrix factorization or the combination of both methods. The results show that the characteristics of the dataset can greatly influence the outcome of the prediction models. By carefully choosing and manipulating the input data, we improve the accuracy of the prediction model by over 30% in some cases.