Comparative Analysis of Graph Neural Networks for Node Regression on Wiki-Squirrel dataset (bachelor's Research Project).
Bachelor's Research Project focused on node regression within graph structures.
Wikipedia article networks dataset named “Wiki-Squirrel dataset” containing three different topics “Chameleon”, “Crocodile”, and “Squirrel”. Conducted different data preprocessing tasks including normalization, outlier handling and one-hot encoding to ensure data quality and consistency.
Implemented and fine-tuned four distinct Graph Neural Network (GNN) models—GCN, GAT, GATv2, and GraphSAGE—exploring diverse architectural approaches for node regression tasks. Additionally, designed and applied four unique loss functions—MSE, RMSE, MAE, and MAPE—to compare their effectiveness.
Executed a comparative analysis to evaluate and compare the performance of GNN models and loss functions across multiple datasets.
- Supervisor: Prof. Mostafa H. Chehreghani
- University: Amirkabir University of Technology (Tehran Polytechnic)
- Semester: Spring 2023
Amirmehdi Zarrinnezhad - amzarrinnezhad@gmail.com
Project Link: https://github.com/zamirmehdi/GNN-Node-Regression