Vol: 34 Page: 279

Authors:
Andrzej Zieliński

Title:
 Estimating missing observations using the EM algorithm applied to transformed data

Language: English

Keywords:
power transformation
multivariate data
missing observations
EM algorithm

Summary:
Estimating missing observations in multidimensional data by means of the EM algorithm works well when the observations are assumed to come from multivariate normal distribution. Otherwise data inserted in the gaps may appear to occur beyond the natural range of considered variables. In the paper the EM algorithm is suggested to be applied to incomplete data having been transformed with the power transformation (instead of unknown exponents one can take the ML estimates). The example shows that due to such an approach one can obtain noticeable improvement in estimating missing observations.