Local Universal Explainer (LUX) -- a rule-based explainer with factual, counterfactual and visual explanations

TL;DR

This work presents Local Universal Explainer (LUX), which is a rule-based explainer that can generate factual, counterfactual and visual explanations and outperforms the existing approaches in terms of simplicity, fidelity, representativeness, and consistency.

Abstract

Explainable artificial intelligence (XAI) is one of the most intensively developed area of AI in recent years. It is also one of the most fragmented with multiple methods that focus on different aspects of explanations. This makes difficult to obtain the full spectrum of explanation at once in a compact and consistent way. To address this issue, we present Local Universal Explainer (LUX), which is a rule-based explainer that can generate factual, counterfactual and visual explanations. It is based on a modified version of decision tree algorithms that allows for oblique splits and integration with feature importance XAI methods such as SHAP. It limits the use data generation in opposite to other algorithms, but is focused on selecting local concepts in a form of high-density clusters of real data that have the highest impact on forming the decision boundary of the explained model and generating artificial samples with novel SHAP-guided sampling algorithm. We tested our method on real and synthetic datasets and compared it with state-of-the-art rule-based explainers such as LORE, EXPLAN and Anchor. Our method outperforms the existing approaches in terms of simplicity, fidelity, representativeness, and consistency.

Figures (17)

• Figure 1: Diagram representing three phases of rule-based explainable model creation. The first phase is focused on selecting and refining representative dataset for local surrogate model creation. The second phase concentrates on generating an oblique, shap-consistent decision tree. The third phase provides visual presentation of local explanations as well as counterfactual generation mechanisms.
• Figure 2: Impact of the generated data on the explanations generated with LORE and EXPLAN. The leftmost plot represents distribution of the original data, the following two plots show distribution changed by the data-generation algorithm used by LORE and EXPLAN. The rightmost plot shows how many phantom branches are generated from such an altered distribution. Such phantom branches may generate non-representative counterfactuals, like these marked in red circles.
• Figure 3: The differences in the local neighborhood defined by different rule-based explainers. The local neighborhood was generated for the instance marked red.
• Figure 4: Decision boundaries of a blackbox model and local approximations of that boundary with LORE, EXPLAN and Anchor. It can be seen that in case of linear boundaries, that are hard to capture by tree-based inequalities the existing explainers produce overcomplicated and non-intuitive explanations.
• Figure 5: Five explanations generated for the same instance and the same dataset. It can be seen that the explanations are different in consecutive runs, making them less usable and less intuitive for users.