The comparison of appropriate methods in imputation of the censored values in the geochemical datasets

This study deals with the imputation methods of censored values in the multivariable geochemical data. Presence of the missing values causes limitation in the use of most of statistical methods, e.g. principle component analysis. Excluding the samples which include missing values bias the results and leads to the loss of information. Due to this, consideration of an appropriate approach to deal with missing values is necessary in the analysis of incomplete datasets. In this paper considering the nature of geochemical data, various approaches for imputing the missing values, which have been suggested in the recent years and are easy to be used in the R statistic software, are introduced. Finally, using the complete dataset of the Zafarghand region, these methods are compared with each other. Results show that the application of the multivariable methods in the imputation and particularly the ilr-EM method is preferable to the other methods.