Performance Evaluation of Extreme Learning Machine with k-Means Techniques for clustering

Abstract

Listen

Translate article

This paper aims to analysis the performance of different activation functions and various number of hidden neurons applied to an integrated Extreme Learning Machine (ELM) k-Means algorithm for clustering. ELM is an emerging and effective learning algorithm which is used to project the data into higher dimensional feature space, then k-means techniques used for clustering in this feature space. In this work we applying three different activation functions such as sigmoidal, radial basis and triangular basis function in ELM feature space were examined. ELM k means with different activation functions and various number of hidden neurons where considered for this analysis and the performance were compared with traditional k means algorithm. The experimental results proves that ELM k-Means gives better accuracy then k-Means algorithms especially the sigmoidal activation function along with ELM k means returns high accuracy with best number of hidden neurons for all four data sets.