Correlation Based Regression Imputation (CBRI) Method for Missing Data Imputation

Abstract

To complete missing values in a dataset is crucial for data mining and machine learning applications. If any parameter of a dataset has missing values, the values of the other parameters corresponding to those missing values should not be excluded from the dataset in order to prevent information in the dataset. Missing values should be handled carefully to avoid their affecting analyses and to prevent loss of information. There are many methods to predict missing values (imputation) that take into account other values of the relevant parameter, but these methods do not consider other parameters. In this study, an algorithm considering other parameters is proposed and its performance is compared with methods that calculate missing data without considering other parameters. The proposed method (CBRI) has been tested with a real dataset, and much more successful results have been obtained compared to the two commonly used imputation methods, mean imputation and median imputation.