Article Preview
TopThere are at least three different ways of dealing with missing data based on Little and Rubin (2002): single imputation, multiple imputation, and iterative procedure.
Single imputation strategies provide a single estimate for each missing data value. Many methods for imputing missing values are single imputation methods, such as, C4.5 algorithm, kNN method, and so on. We can partition single imputation methods into parametric methods and nonparametric ones. The parametric regression imputation methods are superior if a dataset can be adequately modeled parametrically, or if users can correctly specify the parametric forms for the dataset. Non-parametric imputation (Qin et al., 2007) offers a nice alternative if users have no idea on the actual distribution of a dataset because the method can provide superior fits by capturing structure in datasets. While much work focus on modeling data by parametric or nonparametric approaches, in Engle et al. (1986) have studied the semi-parametric model. They model the electricity demand 
as the sum of a smooth function 
of monthly temperature t, and a linear function of 
and 
, as well as 11 monthly dummy variables 
, to build a semi-parametric model firstly. In fact, semi-parametric model is more ordinary in real application than nonparametric model or parametric model because we always contain a little but no all information on our datasets, however, there are a little literatures, such as, Nikulin (2008), focusing on this issue because of the analysis complexity, in this paper, we introduce SIIA algorithm to model the partial parametric model for filling up iteratively missing target values.