An Intelligent Missing Data Imputation Techniques: A Review

Abstract

The incomplete dataset is an unescapable problem in data preprocessing that primarily machine learning algorithms could not employ to train the model. Various data imputation approaches were proposed and challenged each other to resolve this problem. These imputations were established to predict the most appropriate value using different machine learning algorithms with various concepts. Furthermore, accurate estimation of the imputation method is exceptionally critical for some datasets to complete the missing value, especially imputing datasets in medical data. The purpose of this paper is to express the power of the distinguished state-of-the-art benchmarks, which have included the K-nearest Neighbors Imputation (KNNImputer) method, Bayesian Principal Component Analysis (BPCA) Imputation method, Multiple Imputation by Center Equation (MICE) Imputation method, Multiple Imputation with denoising autoencoder neural network (MIDAS) method. These methods have contributed to the achievable resolution to optimize and evaluate the appropriate data points for imputing the missing value. We demonstrate the experiment with all these imputation techniques based on the same four datasets which are collected from the hospital. Both Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are utilized to measure the outcome of implementation and compare with each other to prove an extremely robust and appropriate method that overcomes missing data problems. As a result of the experiment, the KNNImputer and MICE have performed better than BPCA and MIDAS imputation, and BPCA has performed better than the MIDAS algorithm.

Keywords

• data imputation technique
• missing data
• machine learning
• deep learning