Predicting Student Behavior Using a Neutrosophic Deep Learning Model

Abstract

We developed an information system using an object-oriented programming language 
and a distributed database (DDB) consisting of multiple interconnected databases across a computer 
network, managed by a distributed database management system (DDBMS) for easy access. An 
intelligent system was designed to assess the difficulty level of preliminary exams and select 
top-performing advanced students using a Neutrosophic Deep Learning Model. The dataset was 
randomly split into training (80%) and testing (20%) sets, and the model, trained with the Adam 
optimizer at a 0.001 learning rate over 50 epochs, incorporated early stopping based on validation 
loss. This system, implemented at a traditional Egyptian university, achieved a 95% accuracy in 
predicting student dropout. Student behavior, influenced by personal, environmental, and 
academic factors, is often evaluated subjectively, leading to inconsistent results. Traditional machine 
learning approaches struggle with the inherent uncertainty in behavioral data. To address this, we 
combined neutrosophic theory—a mathematical framework that accounts for truth, falsity, and 
indeterminacy—with deep learning, which excels at learning complex data relationships, to predict 
student outcomes such as dropout rates. Evaluating the model on student data, including 
attendance and grades, showed superior accuracy, achieving a determination coefficient of 0.95, 
demonstrating the approach's potential for identifying at-risk students and enabling targeted 
interventions. 

DOI: 10.5281/zenodo.13997076