A Novel K-Nearest Neighbor Technique for Data Clustering using Swarm Optimization.

Abstract

Clustering is the process of grouping a set of physical or abstract objects into classes of similar objects. The k nearest neighbors is selected by using a predefined distance metric (Hamming distance, Euclidean distance etc) to sense the selected similarity metrics. With KNN technique dimensionality reduction is applied to avoid the effects of the curse of dimensionality. The PSO algorithm optimizes the performance of a KNN classifier by finding the best k values for production of the best clustering performance. This paper presents enhanced method of clustering using K-Nearest neighbor with particle swarm optimization (KNN_PSO) over K-Nearest neighbor (KNN) algorithms which can be traceable even for large datasets. The KNN, PSO and KNN_PSO clustering algorithms are analyzed for different datasets using accuracy as the performance measure. The experimental results exhibit that the clustering using K-Nearest Neighbor with PSO approach outperforms K-Nearest Neighbor algorithm with respect to overall accuracy.