Nonparametric K-means algorithm with applications in economic and functional data

Abstract

Inspired by the well-known relationship between K-means algorithm and Expectation-Maximization (EM) algorithm for mixture models, we propose nonparametric K-means algorithm for estimation of nonparametric mixture of regressions and mixture of Gaussian processes. The proposed methods are illustrated by extensive numerical simulations, comparisons, and analysis of two real datasets. Simulation studies and applications demonstrate that our method is an effective and competitive procedure for modified EM algorithm in nonparametric mixture settings.

KEYWORDS:

• K-means algorithm
• mixture models
• nonparametric regression
• Gaussian processes
• EM algorithm

Additional information

Funding

Feng’s research is supported by 2014 Zhejiang Province Philosophy Social Sciences Planning Project (No: 14NDJC231YB) and Shanghai University of Finance and Economics Graduate Student Innovation Fund Project 2014 (No: CXJJ-2014-393).