Comparison of missing values handling techniques using mice package tools of r software and logistic regression model

Abstract

The paper presents the result of the research concerning comparisson of various techniques of missing values multiple imputation by chained equations (MICE) with the use of logistic regression at the stage of the model verification. The presence of missing values in the data complicates the data processing and increases the risk factor in the process of solving various problems in various areas of data science techniques use. The simulation process was performed on the basis of the apply of both R and KNIME software tools. The Mammographic Mass dataset from Machine Learning Repository was used as the experimental data during the simulation process. Implementation of the step-by-step process of missing values handling involved the data analysis and missing values visualization at the first step. Then, we have performed the missing values handling with the use of various techniques which are available in MICE package of R software. The quality of the data processing at each step of this procedure implementation was estimated with the use of logistic regression model based on ROC analysis with calculation of the quantitative criteria: AUC (area under roc curve), Akaike and Bayesian ones. At the final step, we have compared various techniques of missing values handling for purpose of selection from them the best variants taking into account the used criteria.