Graph Neural Networks for Multivariate Time Series Regression with Application to Seismic Data

Authors: Stefan Bloemheuvel, Jurgen van den Hoogen, Dario Jozinovic, Alberto Michelini and Martin Atzmueller

---

Data

The data (too big to host on github itself) can be downloaded at: https://zenodo.org/record/5767221
In the data folder, the input_ci.npy file should be placed
The input_cw.npy file should be placed in data/othernetwork

Requirements

• geopy==2.2.0
• keras==2.8.0
• networkx==2.7.1
• numba==0.56.2
• numpy==1.22.3
• scikit-learn==1.0.2
• scipy==1.8.0
• sklearn==0.0
• spektral==1.1.0
• tensorflow==2.8.0

How to run

Run either main_cnn.py or main_gcn.py with the sys argument 'network1' or 'network2' in terminal, following the with 'nofeatures' or 'main' for the main version. Lastly, a number that serves as the random state for the split.

Example:

$ python main_gcn.py network1 main 1

Here, 'network1' refers to the CI network, 'main' refers to running the main experiment and '1' refers to a seed which can be used.

Results (in MSE)

CI Network

Type	PGA	PGV	PSA03	PSA1	PSA3
SVM	0.36	0.43	0.41	0.37	0.40
KNN	0.32	0.37	0.37	0.35	0.38
XGBoost	0.28	0.32	0.33	0.31	0.33
RF	0.28	0.32	0.33	0.31	0.33
GAT	0.30	0.26	0.26	0.28	0.28
Jozinovic et al.	0.22	0.26	0.24	0.26	0.25
Kim et al.	0.26	0.23	0.23	0.24	0.24
TISER-GCN	0.20	0.21	0.19	0.20	0.21

CW Network

Type	PGA	PGV	PSA03	PSA1	PSA3
GAT	0.49	0.52	0.52	0.49	0.56
SVM	0.43	0.51	0.58	0.51	0.40
KNN	0.45	0.51	0.60	0.53	0.41
XGBoost	0.42	0.48	0.57	0.51	0.39
RF	0.40	0.47	0.56	0.50	0.39
Kim et al.	0.35	0.40	0.38	0.35	0.37
Jozinovic et al.	0.35	0.37	0.35	0.40	0.36
TISER-GCN	0.30	0.30	0.29	0.31	0.33

Cite

If you compare with, build on, or use aspects of this work, please cite the following:

@article{bloemheuvel2022graph,
  title={Graph neural networks for multivariate time series regression with application to seismic data},
  author={Bloemheuvel, Stefan and van den Hoogen, Jurgen and Jozinovic, Dario and Michelini, Alberto and Atzmueller, Martin},
  journal={International Journal of Data Science and Analytics},
  pages={1--16},
  year={2022},
  publisher={Springer}
}