The study of EM algorithm based on forward sampling

Dataset with missing values is quite common in naive bayesian classifier applications, which affects the capability of classifier. And handling missing values has become a research hot issue in the classification field. EM algorithm , a method of iteration , has been widely applied to statistical inferences involving incomplete data such as missing data , censoring data , group data and data bearing disgusting parameters. This paper introduces EM algorithm. To deal with the defects of EM algorithm´s slow convergence speed and local convergence. Forward sampling is introduced into EM algorithm. First, get hold of the swatch using forward sampling; then compute the expectation of missing data in the sample; the expectation is used as the initialization in EM algorithm. Finally, the experiment validates the improved EM algorithm is better than conventionality EM algorithm.