Directly Constructing Multiple Features for Classification with Missing Data using Genetic Programming with Interval Functions

Missing values are a common issue in many industrial and real-world datasets. Genetic programming-based multiple feature construction (GPMFC) is a recent promising filter approach to constructing multiple features for classification using genetic programming (GP). GPMFC has been demonstrated to improve classification performance and reduce the complexity of many decision trees and rule-based classifiers, but it cannot work with missing data. To deal with missing data, this paper propose IGPMFC, an extension of GPMFC that use interval functions as the GP function set to directly construct multiple features for classification with missing data. Empirical results on five datasets and four classifiers show that IGPMFC can substantially improve the performance and reduce the complexity of the classifiers when faced with missing data.