Maximum likelihood constrained Gaussian kernel probability density function estimation using continuous ACO

摘要

Probability density function (PDF) estimation problem is a core stage of the pattern recognition, data analysis and other engineering applications. One of the popular maximum -likelihood (ML) constrained methods for Gaussian kernel PDF estimation is expectation maximization (EM) algorithm. However, it suffers from low convergence speed and also ill positioning of the kernels. We proposed a new PDF estimation method based on the continuous domain ant colony optimization (ACOR) with the aim to maximize the likelihood of data patterns. The proposed approach outperforms the EM algorithm in estimating the Gaussian kernel parameters. One stochastic dataset involved in evaluating the algorithm. Results show that applying modified ACOR provides more accurate estimation of the PDF parameters in lower convergence time. (C) 2015 Elsevier GmbH. All rights reserved.