Accuracy Comparison of Data Imputation Estimation Methods Between Partial Least Squarer of Structural Equation Modeling and K-Nearest Neighbor

Abstract

This study aimed to the accuracy comparison of data imputation estimation methods between partial least squares of structural equation modeling  (PLS of SEM) and k-nearest neighbor (K-NN). The measurement accuracy of the model based on the mean  magnitude of relative error (MMRE). Model development using data from the online database UCI data set waveform database generator. Indicators 21 (1,200 sets) methods were as follows:  1) Data set was divided into 2 groups (experimental group of 1,000 sets and test group of 200 sets ). 2) The experimental group was analyzed by three main factors. 3) PLS of SEM method: Created a SEM with  three main factors, then the remaining factors to created new the relationships with PLS  method and created new SEM. The test data was substituted in the equation to find the MMRE which was 36.90% (accuracy was 63.1%). 4) K-NN method: Selected the main factor was the relationship of the missing data. Measure the Euclidean distance between test group and selected 5 (K=5) of data sets were nearest to the missing data for estimate by mean. The MMRE which was 61.52% (accuracy was 38.48%). Thus, comparing estimates of missing data showed that using the PLS of SEM method were more accuracy about 24.62% and MMRE declined than K-NN method.