Weighting imputation methods and their evaluation under shell-neighbor machine

The paper studies three typical weighting strategies for Shell-Neighbor Imputation (SNI) algorithm, while there are many weighting modes that can be used in the SNI. To best capture the imputation efficiency, a new metrics, called goodess, is proposed for evaluating imputation algorithms. We conduct some experiments for examining the proposed approached, and demonstrate that (1) distance-frequency-weighting strategy is the best one for the shell-neighbor imputation; (2) the goodness is much better than the RMSE if there is a few individual values of serious deviation, otherwise, the goodness is the same as the RMSE at measuring the imputation efficiency.