Explainable Spatial Machine Learning Methods for Complex Socioeconomic Data

The aim of this project is to develop advanced explainable spatial machine learning-based artificial intelligence models and algorithms for clustering and predicting complex spatial data, focusing on problems involving socioeconomic data. These methods help uncover complex spatial patterns and provide a deeper understanding of how socioeconomic factors relate to outcomes across different areas. They are applicable to a wide range of real-world problems in various interdisciplinary fields, including public safety, social sciences, geosciences, and advanced manufacturing. These methods are made accessible and interpretable through visualization software and open-source tools. In this project, innovative, explainable spatial machine models and algorithms are developed, motivated by the analysis of complex spatial socioeconomic data. The clustering method is based on a new Bayesian random graph partition model capable of handling diverse socioeconomic data types and distributions while incorporating prior domain knowledge. A spatial nonparametric regression model utilizing spatially aware ensemble decision tree models and a scalable optimization algorithm also is developed. These methods are well-suited for estimating complex nonlinear effects of socioeconomic covariates on spatial responses and for making predictions at unobserved locations. The algorithm works with both normal and non-normal outcome data and can handle large datasets and high-dimensional covariates. Additionally, the project develops visualization software and a new spatial Shapley value framework to enhance the interpretability of model results. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.