Feature Selection for Capacitated Vehicle Routing Problem Algorithm Selection

The practical application of feature selection is evident in the machine learning domain. The purpose of a learning activity is to identify a mapping of features to output that generalizes the data. Therefore, identifying significant features is an utmost requirement. Feature selection approaches provide a means to decrease computational time, enhance predictive accuracy, and facilitate a deeper comprehension of the underlying data. This work presents an overview of feature selection methodologies documented in the literature. We additionally apply feature selection techniques to a dataset containing capacitated vehicle routing problem instances features and algorithm performance, designed to facilitate algorithm selection. Random forest which is an embedded method of feature selection is used which explicitly provides importance of each attribute for the underlying classification task.