A Survey of Machine Learning-Based Approaches for Missing Value Imputation

Missing values create issues during the analysis of the dataset. Learning algorithms in an asymmetrical dataset can generate an overrated classification accuracy due to a bias towards the majority class at the expense of the minority class. Missing values in the dataset have a negative impact on the imputation of accuracy; therefore, it could lead to a different output. Some algorithms cannot handle missing values properly, while some techniques give efficient results to estimate the missing values. It is very important to handle missing data because many machine learning algorithm performances reduces due to missing values. It might be possible that the original datasets have some missing data for many factors like data were not kept in a file, data had been corrupted, etc. In this paper, some techniques are discussed which are employed to impute the missing data, and these techniques are compared using their merits and drawbacks.