A study of MBR-based spatial access methods: how well they perform in high-dimensional spaces

Many spatial applications deal with data characterized by extremely high dimensionality. Unfortunately, other than the fact that contemporary spatial access methods (SAMs) are inadequate to handle large sets of high dimensional data, we know little about the underlying causes of their inadequacy. The paper investigates the performance of spatial access methods that approximate extended objects by minimum bounding rectangles (MBRs). It exposes the conceptual problems of traditional MBR based SAMs resulting in their inability to handle high dimensional data. The paper shows that a new MBR based structure, called the QSF-tree, which avoids the problems of traditional SAMs, gracefully adapts to accommodate the increasing dimensionality of the spatial data. The experimental evidence demonstrates that QSF-trees outperform three popular MBR based SAMs in both low- and high-dimensional spaces. As the number of dimensions grows, the improvements have a tendency to increase.