Genetic Programming-Based Selection of Imputation Methods in Symbolic Regression with Missing Values
Lists
Only you see this
Abstract
Data incompleteness represents a serious issue in real-world applications of machine learning. Imputation methods are algorithms for restoring missing values in the data based on other available entries. Imputation methods have an influence on the learning performance on incomplete data. Therefore, the choice of a right imputation method has an important role when constructing prediction models. It is common to use one imputation method to impute all the incomplete features. However, the imputation method that works well for some features might not be suitable for others, hence, it would be more useful to select the right imputation method for each feature. In fact, selecting an imputation method for the whole data set is still a challenging issue, let al one selecting different imputation methods for all incomplete features. Therefore, this work proposes the use of genetic programming to search for the right combination of imputation methods for symbolic regression. The role of GP is to select imputation methods for incomplete features and evolve symbolic regression. It incorporates a heterogeneous set of imputation methods as part of the symbolic regression process. The results show that the proposed method can automatically find the most effective combination of imputation methods for a variety of incomplete regression data sets.
BibTeX
@inproceedings{alhelali2020genetic,
author = {Baligh Al-Helali and Qi Chen and Bing Xue and Mengjie Zhang},
title = {{Genetic Programming-Based Selection of Imputation Methods in Symbolic Regression with Missing Values}},
booktitle = {Australasian Conference on Artificial Intelligence},
year = {2020},
url = {https://openreview.net/forum?id=kwdSEIj2Ee},
}Loading related content...