A Supervised Feature Selection Approach Based on Global Sensitivity

In this paper we propose a wrapper method for feature selection in supervised learning. It is based on the global sensitivity analysis; a variancebased technique that determines the contribution of each feature and their interactions to the overall variance of the target variable. First-order and total Sobol sensitivity indices are used for feature ranking. Feature selection based on global sensitivity is a wrapper method that utilizes the trained model to evaluate feature importance. It is characterized by its computational efficiency because both sensitivity indices are calculated using the same Monte Carlo integral. A publicly available data set in machine learning is used to demonstrate the application of the algorithm.