Improvement On Clustering Algorithm Based On Weighted Likelihood

The rapid development of the Internet is constantly promoting the quality of life of human society,accompanied by a huge amount of data,but not all information is needed and valuable.Therefore,it is extremely important to grasp the valuable part of data.The finite mixture model,which can approximate almost all probability densities,is an efficient tool for solving complex distributions and has become a hot research topic.The Gaussian mixture model is the most popular model.Many scholars have conducted research on it and applied it in every aspect of life.Luca Greco and Claudio Agostinelli propose the weighted-CEM algorithm on the basis of Gaussian mixture model in 2018.In the case of a small amount of outliers in the data,the model parameters estimated by the algorithm are very robust,and the accuracy of the algorithm for identifying outliers is also high.However,as the number of outliers increases,the parameters estimated by the algorithm are still not robust enough,and the accuracy of outlier recognition decreases.Therefore,this paper proposes Weighted-DAEM algorithm which is an improved DAEM algorithm.Through weight likelihood function and outlier detection,this algorithm carries out a research on clustering.The algorithm also introduces the lower bound theory of annealing parameter and performs clustering and outlier recognition on Gaussian mixture data with a small amount of outliers.At the same time,this method maintains the convergence of the original algorithm.Experimental results of simulated data and real data show that under certain conditions,the improved DAEM algorithm has better clustering effect on the Gaussian mixture model and more accurate identification of outliers.