Explainable AI via Large Language Models: Translating Neural Network Behavior into Interpretable Decision Trees

Abstract

Deep learning models achieve state-of-the-art performance across various domains but suffer from a lack of interpretability. We propose a novel approach that leverages large language models (LLMs), such as GPT-4, to generate human-readable decision trees that approximate and explain the behavior of complex black-box neural networks. Our method facilitates transparent model auditing and knowledge distillation by transforming neural predictions into decision rules via zero-shot prompting. Experiments on healthcare and finance datasets show that the LLM-generated trees achieve high fidelity with the underlying models while improving interpretability.

Ähnliche Arbeiten