Abstract

In this century access to large and complex datasets is much easier. These datasets are large in dimension and volume, and researchers are interested in methods that are able to handle this type of data and at the same time produce accurate results. Machine learning methods are particularly efficient for this type of data, where the emphasis is on data analysis, and not on fitting a statistical model. A very popular method from this group is Random Forests which have been applied in different areas of study on two types of problems: classification and regression. The former is more popular, while the latter can be applied for data analysis. Moreover, many efficient techniques for missing value imputation were added to Random Forest over time. One of these methods which can handle all types of variables is MissForest. There are several studies that applied different approaches to improve the performance of classification type of Random Forests, but there are not many studies available for regression type. In the present study, it is evaluated if the performance of regression type of Random Forests and MissForests could be improved by applying Genetic Algorithms as an optimization method. The experiments were conducted on five datasets to minimize the mean square error (MSE) of the Random Forest and imputation errors of the MissForest. The results showed the superiority of the proposed method in comparison to the classical Random Forest methods.