ExplainabilityTransparencyAccountabilitycs.LG
On Computing Relevant Features for Explaining NBCs
Yacine Izza, Joao Marques-Silva
July 11, 2022
ENIGMA@KR
6 citations
Key Takeaways
-
Model-agnostic explainable AI tools can produce incorrect explanations, highlighting the need for alternative approaches like formal methods that provide mathematically rigorous explanations rather than relying on general-purpose explanation techniques.
-
Formal explanation methods (PI-explanations) face a practical problem: they generate explanations that are too long and complex to be useful, which this research addresses by computing smaller, more relevant feature sets that maintain accuracy while improving readability.
-
For Naive Bayes Classifiers specifically, computing relevant features is computationally feasible in practice, despite theoretical complexity concerns, making this formal approach potentially viable for real-world applications where simpler models are used.
-
Succinct explanations are achievable without sacrificing precision, as the experiments demonstrate that smaller sets of relevant features can be extracted while maintaining the mathematical rigor that formal XAI methods provide—offering a practical path toward both trustworthy and understandable AI systems.
Abstract
Despite the progress observed with model-agnostic explainable AI (XAI), it is the case that model-agnostic XAI can produce incorrect explanations. One alternative are the so-called formal approaches to XAI, that include PI-explanations. Unfortunately, PI-explanations also exhibit important drawbacks, the most visible of which is arguably their size. The computation of relevant features serves to trade off probabilistic precision for the number of features in an explanation. However, even for very simple classifiers, the complexity of computing sets of relevant features is prohibitive. This paper investigates the computation of relevant sets for Naive Bayes Classifiers (NBCs), and shows that, in practice, these are easy to compute. Furthermore, the experiments confirm that succinct sets of relevant features can be obtained with NBCs.