Spatial Data Partitioning Method Based on Manifold Learning

Spatial data partitioning is a prerequisite for high efficient spatial joins within spatial database systems. Low data redundancy and high data balance rates are difficult to maintain however, using existing spatial data partitioning methods. We propose a spatial data partitioning algorithm based on manifold learning. Manifold learning can retain the structures of source data to construct a data partitioning strategy and mapping method before dimensionality reduction. Assigning neighboring objects to the same data block reduces data redundancy while mapping objects to the smallest data block adds data balance. Experiments show that spatial data partitioning based on manifold learning can reduce the data redundancy rate to very low level with good data balance.