Incremental Shared Nearest Neighbor Density-Based Clustering Algorithms for Dynamic Datasets

Dynamic datasets undergo frequent changes where small number of data points are added and deleted. Such dynamic datasets are frequently encountered in many real world applications such as search engines and recommender systems. Incremental data mining algorithms process these updates to datasets efficiently to avoid redundant computation. Shared nearest neighbor density based clustering (SNN-DBSCAN) is a widely used clustering algorithm, mainly for its robustness. Existing incremental extension to SNNDBSCAN cannot handle deletions to dataset and handles insertions only point by point. We overcome both these bottlenecks by efficiently identifying affected parts of clusters while processing updates to dataset in batch mode. We present three different incremental algorithms with varying efficiency at elimination of redundant computation. We show effectiveness of our algorithms by performing experiments on large synthetic as well as real world datasets. Our algorithms are up to 2 Orders of Magnitude faster than nonincremental algorithm and up-to 50 times faster than existing incremental algorithm while guaranteeing exact same output.