A Review of Missing Data Handling Techniques for Machine Learning

Abstract

Real-world data are commonly known to contain missing values, and consequently affect the performance of most machine learning algorithms adversely when employed on such datasets. Precisely, missing values are among the various challenges occurring in real-world data. Since the accuracy and efficiency of machine learning models depend on the quality of the data used, there is a need for data analysts and researchers working with data, to seek out some relevant techniques that can be used to handle these inescapable missing values. This paper reviews some state-of-art practices obtained in the literature for handling missing data problems for machine learning. It lists some evaluation metrics used in measuring the performance of these techniques. This study tries to put these techniques and evaluation metrics in clear terms, followed by some mathematical equations. Furthermore, some recommendations to consider when dealing with missing data handling techniques were provided.