Measuring What LLMs Think They Do: SHAP Faithfulness and Deployability on Financial Tabular Classification

This research investigates the critical gap between Large Language Models' self-reported feature importance and actual SHAP values in financial classificatio...

Level: advanced

By Saeed AlMarri, Mathieu Ravaut, Kristof Juhasz, Gautier Marti, Hamdan Al Ahbabi, Ibrahim Elfadel

Category: research