Educational Data Mining Techniques for Student Performance Prediction: Method Review and Comparison Analysis.

Yupei Zhang,Jiaqi Cui,Rui An,Huan Dai,Yue Yun,Xuequn Shang

doi:10.3389/fpsyg.2021.698490

Abstract

Student performance prediction (SPP) aims to evaluate the grade that a student will reach before enrolling in a course or taking an exam. This prediction problem is a kernel task toward personalized education and has attracted increasing attention in the field of artificial intelligence and educational data mining (EDM). This paper provides a systematic review of the SPP study from the perspective of machine learning and data mining. This review partitions SPP into five stages, i.e., data collection, problem formalization, model, prediction, and application. To have an intuition on these involved methods, we conducted experiments on a data set from our institute and a public data set. Our educational dataset composed of 1,325 students, and 832 courses was collected from the information system, which represents a typical higher education in China. With the experimental results, discussions on current shortcomings and interesting future works are finally summarized from data collections to practices. This work provides developments and challenges in the study task of SPP and facilitates the progress of personalized education.

Highlights

Educational data mining (EDM), a very young research field, focuses on learning latent patterns in various educational situations, including student’s knowledge analysis (Yeung and Yeung, 2018), student’s learning behavior analysis (Juhanák et al, 2019), teacher’s curriculum planning (Reeves, 2018), course time arrangement (Zhang et al, 2018a)
The prerequisite course grades play the most important role in the random forest classifier, which shows that the records might uncover the characters of a student on learning
The performance of all models is significantly improved with integrating the features from the prerequisite courses (p-value < 0.01)

Summary

Introduction

Educational data mining (EDM), a very young research field, focuses on learning latent patterns in various educational situations, including student’s knowledge analysis (Yeung and Yeung, 2018), student’s learning behavior analysis (Juhanák et al, 2019), teacher’s curriculum planning (Reeves, 2018), course time arrangement (Zhang et al, 2018a). While several review papers have summarized previous EDM research studies (Shahiri and Husain, 2015; Saa, 2016), this paper provides a more completed survey on the problem of SPP from the perspective of machine learning and data mining. SPP could help to check the curriculum program and to optimize the course system (Reeves, 2018). The data-driven SPP study provides an objective reference for the education system

Objectives

Methods

Findings

Conclusion