Improving on Coalitional Prediction Explanation

Abstract

Machine learning has proven increasingly essential in many fields but a lot obstacles still hinder its use by non-experts. The lack of trust in the results obtained is foremost among them, and has inspired several explanatory approaches in the literature. These approaches provide a great insight on the predictions of a model, but at a cost of a long computation time. In this paper, we aim to further improve the detection of relevant attributes influencing a prediction, on the strength of feature selection methods.