Personalized Explainable Transformer Models for Student Performance Prediction

Title

Personalized Explainable Transformer Models for Student Performance Prediction

Creator

Jayaraj, Jishnu Plavinchottil; Ashok Immanuel, V.; Chandra, J.

Description

The research presents a unified framework for forecasting student academic achievement using a transformer-based architecture supported by Explainable Artificial Intelligence techniques. The research is motivated by the need to combine predictive accuracy of transformer models with interpretability in student performance prediction. The framework applies an adapted Feature Tokenizer Transformer to the UCI Student Performance dataset and integrates SHAP and LIME methods to generate instance level, human readable explanations for each prediction. These explanations can help educators design targeted interventions. A Random Forest regressor is included as a baseline for comparison. The experiment results showed that the Random Forest performed slightly better than the Feature Tokenizer Transformer, which could be due to the small size of the dataset and certain features having a strong impact on the results. Nevertheless, the results show that modern deep learning models combined with personalized explainability offer a practical foundation for scalable solutions in more complex educational datasets, which helps connect high performing prediction models with actionable insights, contributing to the development of interpretable, data driven student support systems. 2025 IEEE.

Source

Proceedings - 2025 International Conference on Transformative Computing Technologies, ICTCT 2025;pp.44-49

Date

01-01-2025

Publisher

Institute of Electrical and Electronics Engineers Inc.

Subject

Explainable AI; LIME; SHAP; Student Performance Prediction; Transformer

Coverage

Jayaraj J.P., CHRIST University, Department of Computer Science, Bangalore, India; Ashok Immanuel V., CHRIST University, Department of Computer Science, Bangalore, India; Chandra J., CHRIST University, Department of Computer Science, Bangalore, India

Rights

Restricted Access; Hardcopy may be available in the library

Relation

ISBN: 979-833159195-3;

Format

online

Language

English

Type

Conference paper

Identifier

https://doi.org/10.1109/ICTCT69201.2025.00026

https://www.scopus.com/pages/publications/105037468909?origin=resultslist