Evolutionary computation for feature manipulation: Key challenges and future directions

Bing Xue, Mengjie Zhang

Abstract: In machine learning and data mining, feature manipulation is a data pre-processing step to increase the quality of a feature space, which can significantly improve the performance of a learning algorithm in terms of the accuracy, the learning speed, and the complexity and the interpretability of the learnt models. However, feature manipulation is a difficult task and facing more challenges along with the trend that more and more data is collected in many domains. Evolutionary computation (EC) techniques have recently attracted much attention for dealing with complex feature manipulation problems. Current work has demonstrated some strengths of EC for feature manipulation, but also shown some limitations and issues that need to be addressed. More importantly, there are some highly interesting research topics in the EC for feature manipulation area, which could potentially result in promising approaches to data analysis in a variety of real-world applications. This position paper describes and discusses the main issues and key challenges of feature manipulation, and also provides a number of directions for further consideration in future research.