Combining instance selection for better missing value imputation

In practice, the data collected from data mining usually contain some missing values. Imputation is the process of replacing the missing values in incomplete datasets. It is usually based on providing estimations for missing values by reasoning from the observed data. Consequently, the effectiveness of missing value imputation is heavily dependent on the observed data (or complete data) in the incomplete datasets. The objective of this study is to investigate the effect of performing instance selection to filter out some noisy data (or outliers) from a given dataset on the imputation task. Specifically, four different processes for combining instance selection and missing value imputation are proposed and compared in terms of data classification. The experimental results based on 29 datasets containing categorical, numerical, and mixed attribute types of data show that the process of performing instance selection first and imputation second allows the k-NN and SVM classifiers to outperform the other processes over the categorical and numerical datasets. For the mixed type of datasets, k-NN performs the best when instance selection is performed again on the datasets produced by the second process. Finally, some specific decision rules about when to employ which process are also provided for future research.