Spatial objects classification using machine learning and spatial walk algorithm

Abstract

This article presents a novel method for classifying spatial objects by learning node representations via a spatial walk algorithm. The findings show that considering both the attributes of objects and their topological relationships enables more efficient and precise spatial objects’ classification than methods that only consider the objects’ characteristics. The method emphasizes the importance of spatial dependencies in learning representations for spatial data. A distinctive feature of the method is its focus on local analysis of the neighborhood structure of the node under investigation. The spatial walk algorithm offers a defined path generation scheme, facilitating a deeper understanding of local spatial dependencies between objects. This approach provides a more accurate representation of the essential relationships between spatial objects than random path generation and enhances the classification results, as demonstrated in three different classification scenarios. The method proves particularly effective in the context of spatial objects, where proximity and a limited number of neighbors play a significant role. This is exemplified in the classification of planning areas in spatial development plans.

Keywords

• random walk
• graphs
• neural networks
• representation learning
• geospatial data