.png){kind=icon}
.png){kind=icon}
Return
.png){kind=icon}
.png){kind=icon}
.png){kind=icon}
.png){kind=icon}
.png){kind=icon}
Abstract
En 中文
Machine learning classification models are increasingly supporting human experts in many fields, including education, medicine, and finance. However, the complexity of classification models, particularly deep neural networks, makes it difficult for experts to understand their inner workings. By providing transparent and comprehensible explanations of model predictions, Explainable Artificial Intelligence (XAI) aims to solve this problem, thereby building users trust. The notable post-hoc XAI method, Model Understanding through Subspace Explanations (MUSE), attempts to interpret black-box models by analyzing their behaviors in different subspaces. However, it has some limitations, primarily the poor interpretability of instance-wise explanations. To tackle this, we propose an explanation method named Closed Frequent Itemsets Explanation (CFIE). By extracting closed frequent itemsets from data, CFIE generates compact rule-based explanations that enhance instance and class interpretability. Our approach improves interpretability while maintaining high fidelity compared to MUSE. Experimental evaluation on a real dataset demonstrates that CFIE generates more interpretable explanations than state-of-the-art approaches of enhanced fidelity-interpretability trade-off.
Keywords:
Explainable Artificial Intelligence
Post-hoc Explanation
Rule-based Explanation
Closed Frequent Itemsets
Journal
A
IF:
0
Papers:
33
Citations:
0