[Paper Review] Machine Learning Based Student Grade Prediction: A Case Study

Zafar Iqbal, Junaid Qadir+2 more|arXiv (Cornell University)|Aug 17, 2017

Data Stream Mining Techniques24 references86 citations

TL;DR

The paper compares collaborative filtering, matrix factorization (SVD and NMF), and Restricted Boltzmann Machines to predict ITU Electrical Engineering students' course grades using real-world data, with RBM performing best.

ABSTRACT

In higher educational institutes, many students have to struggle hard to complete different courses since there is no dedicated support offered to students who need special attention in the registered courses. Machine learning techniques can be utilized for students' grades prediction in different courses. Such techniques would help students to improve their performance based on predicted grades and would enable instructors to identify such individuals who might need assistance in the courses. In this paper, we use Collaborative Filtering (CF), Matrix Factorization (MF), and Restricted Boltzmann Machines (RBM) techniques to systematically analyze a real-world data collected from Information Technology University (ITU), Lahore, Pakistan. We evaluate the academic performance of ITU students who got admission in the bachelor's degree program in ITU's Electrical Engineering department. The RBM technique is found to be better than the other techniques used in predicting the students' performance in the particular course.

Motivation & Objective

• Motivate use of machine learning to predict student grades for early intervention and retention.
• Evaluate multiple ML techniques (CF, MF, RBM) on real-world ITU student data.
• Develop a feedback model to identify weak students and provide course-domain level insights.
• Propose a forecasting approach that supports instructor interventions based on predicted GPA.

Proposed method

• Formulate grade prediction as a recommender-system problem using a Student-Course-GPA matrix.
• Apply UBCF, SVD, and NMF to predict missing GPAs.
• Utilize RBM as an unsupervised predictor to impute missing grades.
• Center predictor variables by subtracting course average GPA to improve model performance.
• Implement a feedback model linking predicted GPA to course-domain knowledge and instructor alerts.

Experimental results

Research questions

• RQ1 Can collaborative filtering, matrix factorization, and RBM accurately predict next-term grades for ITU students?
• RQ2 Which ML technique yields the most accurate GPA predictions on sparse student-course data?
• RQ3 How can predicted GPAs be translated into actionable feedback for instructors and students?
• RQ4 Does incorporating domain-based knowledge and a learning model improve early warning signals for at-risk students?

Key findings

• RBM performed best among the techniques studied for predicting student performance.
• The ITU dataset comprises 225 students and 24 courses, with 5400 potential cell entries and 1736 observed grades (density 32.14%).
• Pre-admission factors like entry test and HSSC show positive correlations with CGPA (r = 0.29 and r = 0.28 respectively).
• CF, MF (SVD/NMF) offer competitive approaches for sparse data and can be used to predict next-term grades.
• The study provides a feedback model to alert instructors when a student’s predicted GPA in a course is below a threshold (2.67).
• The paper discusses a fitting approach for hidden Markov modeling to infer student knowledge for course domains.