A Comparative Study of LIME and SHAP for Enhancing Trustworthiness and Efficiency in Explainable AI Systems
Trisna Ari Roshinta, Szűcs Gábor
Abstract
In recent years, artificial intelligence (AI) has emerged as a solution in almost every domain, significantly improving our lives. AI’s ability to efficiently extract important features and find valuable patterns from large amounts of data enables faster task completion. However, concerns have been raised about the opacity of decision-making processes in ‘black box’ AI models. This has led to the development of explainable AI (XAI), which aims to make AI systems more transparent and understandable to users. Several XAI methods have been developed with different approaches and outputs. This paper aims to compare the state-of-the-art explainability approaches, with a particular analysis of methods based on feature importance, such as the well-known Local Interpretable Modelagnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) methods. Experiments are conducted with several machine learning methods on two datasets. This paper presents a comparison of LIME and SHAP, which can be used as a reference in selecting an XAI method according to data characteristics and explanation goals. LIME excels in performing local explanations through interpretable models, making it relatively easy to implement. However, LIME explanations are not very stable across tests due to instance perturbations. In contrast, SHAP provides consistent explanations based on Shapley game theory by computing Shapley values that ensure fair feature contributions across the dataset. Despite its computational overhead, SHAP offers more robust and global insights compared to LIME.