A Graph-Based Neural Network Approach to Student Performance Prediction using Mobile Usage Data

• Published in: Indian journal of science and technology Vol. 18; no. 31; pp. 2555 - 2568
• Main Authors: Belina, R Ruth, Beena, T Lucia Agnes
• Format: Journal Article
• Language: English
• Published: 23.08.2025
• ISSN: 0974-6846; 0974-5645
• DOI: 10.17485/IJST/v18i31.1034

Objectives: To obtain enhanced accuracy and reduced specificity in predicting student academic performance based on mobile phone usage. Methods: This paper presents a novel technique called Canberra Normalized Elastic Net Regression-based Stochastic Gradient Multilayer Perceptron Neural Network (CNER-SGMPNN). A student records are collected from the real time dataset. After that, preprocessing is carried out by using Canberra Normalization (CN) technique to remove duplicate data. Followed by, feature selection is performed using Elastic Net Regression (ER) to choose most relevant attributes. Finally, Hamann Indexive SGMPNN framework is employed for student academic performance prediction by measuring correlations between mobile phone usage and academic results. Findings: The results of the analysis demonstrate that the CNER-SGMPNN technique improves accuracy and sensitivity by 8%, while achieving reductions of 39% in specificity, 47% in RMSE, and 48% in MAPE when compared to the existing methods. Novelty: Multilayer Perceptron Neural Network is employed to predict students’ performance with several layers. Hamann similarity index, maxout activation function, and stochastic gradient function are applied to this Neural Network. The Hamann similarity index enables in CNER-SGMPNN technique to meaningfully process binary-encoded behavioral data (like mobile usage data) and students’ performance. This enhances the model’s ability to learn significant relationships between student mobile usage behavior and academic performance. This helps to accurately predict the low academic and high academic performance students. The maxout activation function in Neural Network accurately predicts a low academic performance student by choosing the maximum value of Hamann similarity index outputs. The stochastic gradient function is employed for updating the hyperparameter (i.e. weight) of the Multilayer Perceptron Neural Network to minimize the error in the low academic performance prediction. Keywords: Academic performance prediction, Attribute selection, Data normalization, Mobile phone usage, Multilayer perceptron neural network