ALGORITMA SHARED NEAREST NEIGHBOR BERBASIS DATA SHRINKING

Abstract

Shared Nearest Neighbor (SNN) algorithm constructs a neighbor graph that uses similarity between data points based on amount of nearest neighbor which shared together. Cluster obtained from representative points that are selected from the neighbor graph. The representative point is used to reduce number of clusterization errors, but also reduces accuracy. Data based shrinking SNN algorithm (SSNN) uses the concept of data movement from data shrinking algorithm to increase accuracy of obtained data shrinking. The concept of data movement will strengthen the density of neighbor graph so that the cluster formation process could be done from neighbor graph components which still has a neighbor relationship. Test result shows SSNN algorithm accuracy is 2% until 8% higher than SNN algorithm, because of the termination of relationship between weak data points in the neighbor graph is done slowly in several iteration. However, the computation time required by SSNN algorithm is three times longer than SNN algoritm computational time, because SSNN algorithm constructs neighbor graph in several iteration.