Regression trees to optimize Hot Deck imputation

Abstract

Faced with the persistent need to have a complete database and the search to improve the classic techniques for estimating missing data, the research presents a proposed methodology for the imputation of the loss or lack of data, by combining a Segmentation analysis, specifically a Regression Tree and the classic sequential Hot Deck imputation technique. The estimators for the mean, total and variance are calculated, as well as the empirical validation of the proposal, in which unbiased estimators were obtained. The technique is considered to improve the robustness between losses of 5 and 30% of the data, the variability of the data and the relationships between the variables are maintained. Improve estimates over Hot Deck without the use of segmentation.