Improve Search Performance for R+ tree Learned Spatial Index

Abstract

The R⁺ tree algorithm is used to give index to each spatial data sets of (point, line polygon spatial objects). Our map datasets contain 1048575 records for point, as well as for line and polygon each. Machine learning approach used to learn the indices. Then nearest neighbor queries (NNQ) are carried out in purpose of evaluating system performance for proposed learned index. The evaluation of proposed indexing done based on query execution time. Execution times of k-Nearest neighbor query in   learned methods for all kind of spatial objects (point, line, and polygon) are calculated, that 25 random sample points were took for each of them. In order to compare learnt and conventional indexing, we also developed the traditional indexing using the R+ tree technique. So, a significant finding that goes beyond the suggested approach (R⁺ Learned Spatial Index (R⁺ LSI)) is that we reached less execution times in learned index method for nearest neighbor queries for all three spatial objects. Getting benefits of Microsoft SQL Server database for handling spatial objects indices. We propose that machine learning (ML) allows for the independent creation of specific index structures by making it possible to design a model that reproduce the patterns in the data. We explore the extent to which learning models may replace traditional index structures like R⁺ tree. Finally measuring the learned index (R⁺ LSI) is done by getting benefits of both   parameters which are, mean square error (MSE) and R-square (R²). The estimator is linear regression which is used in machine learning approach to make a model. Enhancing spatial query performance is done through the minimum execution time of spatial NNQ in this research.