Using ML and Graph Convolution Network Models to Identify At-Risk Students by Extracting
The study "Extracting topological features to identify at-risk students using machine
learning and graph convolutional network models" aims to develop a dependable and
precise machine learning approach for predicting students' academic performance and
identifying those at risk of subpar academic outcomes. The study proposes a method
that incorporates graph theory and topological features to obtain a more profound
understanding and reveal structural correlations within the data.
The methodology used in the study involved converting tabular data into graphs using
distance measures such as Euclidean and Cosine to assess similarities between students'
data and construct a graph. The researchers then extracted graph topological features
(GF) from the graph to enrich their data and reveal structural correlations. These
topological features were merged with the original tabular dataset, and various feature
selection methods were utilized to identify the most impactful features in the enhanced
The students were categorized into three classes based on the final dataset attributes
- good, at-risk, or failed. To compare the performance of conventional machine learning-based
models, the study implemented a graph-based convolutional network (GCN). Furthermore,
the researchers developed a knowledge graph using the features of the proposed dataset
to examine the relationships among different features. The proposed method was designed
to identify at-risk students using knowledge graphs and conventional machine learning
The team was led by Professor Nazar Zaki, a specialist in computer science and machine
learning, and included Dr. Balqis Albreiki, an expert in machine learning, data mining,
and graph theory, and Dr. Tetiana Habuza, an expert in machine learning, medical image,
and clinical data analysis.
The findings of this study have significant implications for improving student success
rates and reducing low performance rates in higher education. By accurately identifying
at-risk students early on, interventions and support systems can be put in place to
improve their chances of success. Additionally, the use of graph theory and topological
features provide valuable insights into the factors that contribute to academic performance,
enabling targeted interventions to address specific challenges faced by students.
The multidisciplinary collaboration among experts in machine learning, graph theory,
and education allowed for a comprehensive approach to the research, making important
contributions to the field of educational technology.