A New Clustering Algorithm Based on Non-Negative Matrix Factorization Approach

This paper presents a new clustering algorithm that is based on non-negative matrix factorization approach. The proposed algorithm is executed in two steps. The first step uses non-negative matrix factorization approach for dimensionality reduction to scale-back the computational burden and noise. The second step performs clustering by using the matrix with reduced dimensions obtained during the step 1.The algorithm is compared with two well-known clustering algorithms namely K-means algorithm and hierarchical clustering algorithm. IRIS dataset is used to compare the three algorithms. The algorithms are compared for the different initial values of parameters associated with clustering algorithms, and by presenting dataset with different order to clustering algorithms. The results indicate that the proposed algorithm produces good clusters while addressing some of the issues related to clustering.