Hyperparameter optimization of neural network-driven spatial models accelerated using cyber-enabled high-performance computing

ABSTRACT

Artificial neural networks (ANNs) have been extensively used for the spatially explicit modeling of complex geographic phenomena. However, because of the complexity of the computational process, there has been an inadequate investigation on the parameter configuration of neural networks. Most studies in the literature from GIScience rely on a trial-and-error approach to select the parameter setting for ANN-driven spatial models. Hyperparameter optimization provides support for selecting the optimal architectures of ANNs. Thus, in this study, we develop an automated hyperparameter selection approach to identify optimal neural networks for spatial modeling. Further, the use of hyperparameter optimization is challenging because hyperparameter space is often large and the associated computational demand is heavy. Therefore, we utilize high-performance computing to accelerate the model selection process. Furthermore, we involve spatial statistics approaches to improve the efficiency of hyperparameter optimization. The spatial model used in our case study is a land price evaluation model in Mecklenburg County, North Carolina, USA. Our results demonstrate that the automated selection approach improves the model-level performance compared with linear regression, and the high-performance computing and spatial statistics approaches are of great help for accelerating and enhancing the selection of optimal neural networks for spatial modeling.

KEYWORDS:

• Artificial neural network
• hyperparameter optimization
• spatially explicit modeling
• high-performance computing
• spatial statistics

Acknowledgments

The authors would like to thank three anonymous reviewers for their insightful comments and suggestions. The authors are grateful for the support from the 2^nd International Symposium of Spatiotemporal Computing travel reward. The authors also thank Michael Desjardins for his kind help.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. Effective sample size (aka, adequate sample size) is a notion defined for a sample that is statistically significant and the observations in the sample are correlated.

Additional information

Notes on contributors

Minrui Zheng

Minrui Zheng is a Ph.D. candidate at Department of Geography and Earth Sciences, University of North Carolina at Charlotte. Her research interests include spatial analysis and modeling, cyberinfrastructure and high-performance computing, and machine learning algorithms.

Wenwu Tang

Wenwu Tang is an associate professor at Department of Geography and Earth Sciences, University of North Carolina at Charlotte. His research interests include spatial analysis and modeling, agent-based models and spatiotemporal simulation, cyberinfrastructure and high-performance computing, complex adaptive spatial systems, and land use and land cover change.

Xiang Zhao

Xiang Zhao is a faculty of School of resource and environmental sciences, Wuhan University. His research interest includes land use planning, agent-based modeling and cyberinfrastructure and high performance computing.