Towards efficient imputation by nearest-neighbors: A clustering-based approach

Abstract

This paper proposes and evaluates a nearest-neighbor method to substitute missing values in ordinal/continuous datasets. In a nutshell, the K-Means clustering algorithm is applied in the complete dataset (without missing values) before the imputation process by nearest-neighbors takes place. Then, the achieved cluster centroids are employed as training instances for the nearest-neighbor method. The proposed method is more efficient than the traditional nearest-neighbor method, and simulations performed in three benchmark datasets also indicate that it provides suitable imputations, both in terms of prediction and classification tasks. © Springer-Verlag Berlin Heidelberg 2004.