Measuring What LLMs Think They Do: SHAP Faithfulness and Deployability on Financial Tabular Classification
Saeed AlMarri, Mathieu Ravaut, Kristof Juhasz, Gautier Marti, Hamdan Al Ahbabi, Ibrahim Elfadel
TL;DR
This study evaluates zero-shot LLMs for financial tabular classification by comparing SHAP-based feature attributions with LLM self-explanations and with LightGBM. It finds that LLM self-explanations often misalign with SHAP values and that LLM SHAP attributions diverge from LightGBM, indicating limited faithfulness of explanations. Predictive gains from zero-shot prompting are modest and highly sensitive to class imbalance, raising deployability concerns for risk-sensitive finance. The work suggests mitigation strategies, calibration, and pursuing few-shot or hybrid approaches to improve usefulness and governance in practical applications.
Abstract
Large Language Models (LLMs) have attracted significant attention for classification tasks, offering a flexible alternative to trusted classical machine learning models like LightGBM through zero-shot prompting. However, their reliability for structured tabular data remains unclear, particularly in high stakes applications like financial risk assessment. Our study systematically evaluates LLMs and generates their SHAP values on financial classification tasks. Our analysis shows a divergence between LLMs self-explanation of feature impact and their SHAP values, as well as notable differences between LLMs and LightGBM SHAP values. These findings highlight the limitations of LLMs as standalone classifiers for structured financial modeling, but also instill optimism that improved explainability mechanisms coupled with few-shot prompting will make LLMs usable in risk-sensitive domains.