View Proposal

---

Proposer

Zi Hau Chin

Title

Explainable Machine-Learning Models for Predicting Students at Risk of Course Failure

Goal

Automatic Classification and Prioritisation of Student Support Requests Using Machine Learning

Description

This project involves developing an explainable machine-learning system that identifies students who may be at risk of failing a course. The system will analyse relevant learning information, such as assessment performance, submission patterns, attendance, quiz attempts, and engagement with online learning activities. The student will prepare the dataset, engineer suitable features, and implement several classification models, such as logistic regression, decision trees, random forests, gradient-boosted trees, or explainable boosting machines. The system must provide more than a risk category or probability. It should explain the main factors contributing to each prediction using techniques such as model coefficients, decision paths, feature importance, SHAP, or counterfactual explanations. The final prototype should include a lecturer-facing dashboard that presents the overall risk distribution across a class, individual student predictions, important contributing factors, changes in risk over time, and appropriate warnings about model uncertainty and limitations. The project will also investigate how early reliable predictions can be produced by comparing model performance at different stages of a course, such as Weeks 2, 4, 6, and 8. Models will be evaluated using precision, recall, F1 score, ROC-AUC, PR-AUC, false-positive rate, probability calibration, fairness, and explanation stability. The completed system should be presented as a human-in-the-loop decision-support tool for early academic intervention. It must not be used to make automatic academic, disciplinary, or progression decisions. Which model provides the best balance between predictive performance and interpretability? -How early can course-failure risk be predicted reliably? -Are explanations stable and actionable? -Does predictive performance differ across relevant student groups? -How does prediction change when earlier assessment grades are excluded?

Resources

UCI Student Performance Dataset, OULAD, Student-Performance-Prediction-ML-Comparison

Background

Python, data analysis, supervised machine learning and classification, Data preprocessing and feature engineering, explainable AI

Url

Difficulty Level

High

Ethical Approval

Full

Number Of Students

1

Supervisor

Zi Hau Chin

Keywords

learning analytics, early-warning system, explainable ai, counterfactual explanation, imbalanced classification, calibration, algorithmic fairness, educational data mining

Degrees

Bachelor of Science in Computing Science