An Unsupervised Feed Forward Neural Network Method for Efficient Clustering

This paper presents a Real Unsupervised Feed Forward Neural Network (RUFFNN) clustering method with one epoch training and data dimensionality reduction ability to overcome some critical problems such as low training speed, low accuracy as well as high memory complexity in this area. The RUFFNN method trains a code book of real weights by utilizing input data directly without using any random values. The Best Match Weight (BMW) vector is mined from the weight codebook and consequently the Total Threshold (TT) of each input data is computed based on the BMW. Finally, the input data are clustered based on their exclusive TT. For evaluation purposes, the clustering performance of the RUFFNN was compared to several related clustering methods using various data sets. The accuracy of the RUFFNN was measured through the number of clusters and the quantity of Correctly Classified Nodes.The superior clustering accuracies of 96.63%, 96.67% and 59.36% were for the Breast Cancer, Iris and Spam datasets from the UCI repository respectively. The memory complexity of the proposed method was O(m.n.sm) based on the number of nodes, attributes and size of the attribute.