Iterative KNN imputation based on GRA for missing values in TPLMS

TL;DR: An iterative KNN imputation method which associates with weighted k nearest neighbor (KNN) imputation and the grey relational analysis (GRA) and the experimental results suggest that the proposed method gets a better performance than other methods in terms of imputation accuracy and convergence speed.

Abstract: Missing values are an unavoidable problem in a number of real world applications and how to impute these missing values has become a challenging issue in industrial production. Even though there are some popular imputation methods proposed, these methods perform poorly in the estimation of missing values in the trash pickup logistics management system (TPLMS). The problem of missing values in the TPLMS is significant and may result in unserviceable decision-making. Thus this paper introduces an iterative KNN imputation method which associates with weighted k nearest neighbor (KNN) imputation and the grey relational analysis (GRA). This method is an instance based imputation method that takes advantage of the correlation of attributes by using a grey relational grade instead of Euclidean distance or other similarity measures to search k-nearest neighbor instances. The plausible values for the missing values are estimated from these nearest neighbor instances iteratively. In addition, the iterative imputation allows all available values including the attribute values in the instances with missing values and the imputed values from previous iteration to be utilized for estimating the missing values. Specifically, the imputation method can fill in all the missing values with reliable data regardless of the missing rate of the TPLMS dataset. We experiment our proposed method on several TPLMS datasets at different missing rates in comparison with some existing imputation methods. The experimental results suggest that the proposed method gets a better performance than other methods in terms of imputation accuracy and convergence speed.