Publication Date
Spring 2026
Degree Type
Master's Project
Degree Name
Master of Science in Computer Science (MSCS)
Department
Computer Science
First Advisor
Navrati Saxena
Second Advisor
William Andreopoulos
Third Advisor
Pooja Shyamsundar
Keywords
Graph Neural Networks, VGAE, Link Prediction, O-RAN, Telecom Graphs
Abstract
Currently, mobility management in Open Radio Access Network (O-RAN) networks mostly uses predictive techniques to address issues related to handover failure and optimization of radio resources. Recent studies highlight that the application of Graph Neural Networks (GNNs) in the context of mobile network analytics is ideal. For example, when Variational Graph Autoencoders (VGAEs) are applied to link prediction tasks associated with next-cell selection GNNs are used. Specifically, next cell selection is considered as a link prediction task using a bipartite user-cell graph generated from the Huawei RW deployment dataset with additional edge feature information. In this case, the proposed VGAE architecture yields an accuracy and F1-score of 0.9127 and 0.9190 on the 10k-UE RW graph compared to the baseline results of 0.80 and 0.83, respectively.
Recommended Citation
Sathe, Mugdha Arun, "Improved Variational Graph Autoencoder for Link Prediction on Real-World Telecom Graphs" (2026). Master's Projects. 1824.
DOI: https://doi.org/10.31979/etd.5b5v-a8tf
https://scholarworks.sjsu.edu/etd_projects/1824
