FEATURE INFORMATIVE STABLE USER CENTRIC LIME

In Today’s era, The Complexity of Machine learning models keeps raising hence the need forgood Explainable artificial intelligence techniques required. This research describes an improvedversion of Local Interpretable Model Agnostic Explanations (Lime), in mitigating inconsistency issuesin explainations related to feature relevance discrepancies in AI algorithms. The proposed framework,Feature Informative, Stable and User Centric LIME, integrates essential components like LIME withMutual Information (MI) and User centric Category based Explanations for substantially improveXAI Explaination capabilities. Enhanced number of sampling size is used within Feature Informative,Stable and User Centric LIME, thus enabling the invariance to produce explanations that are locallyconsistent. This will make the explanation robust under variations in input data. IncorporatingMutual Information further helps in quantifying the relevance of each feature concerning the model’soutput. It becomes more informative and more accessible to interpret concerning the model’s decisionprocess. Among the most prominent achievements of this research is implementing user Categorybased explanations in Feature Informative, Stable and User Centric LIME. Feature Informative,Stable and User Centric LIME recognizes this heterogeneity within the level of expertise and userrequirements. Therefore, it explains the explanations for experts, intermediates, or laypersons.Experts will receive technically detailed descriptions, with insights under the hood and statisticaljustifications. Intermediates will be served explanations that balance technical depth with intuitiveunderstanding. Laypersons are given simplified explanations that eschew low-level concepts andeasy-to-understand interpretations.This categorization will make explanations accessible and meaningful to a large audience, hencemagnifying the utility and impact of the XAI system. Experimentation on loan approval dataset andmodels of machine learning demonstrates the efficacy of the Feature Informative, Stable and UserCentric LIME framework with significant improvements in consistency, relevance, and robustness ofexplanations in comparison with the traditional LIME. In brief, this research contributes to the novelenhancement of LIME by merging Mutual Information with categories of users for explanationsFeature Informative, Stable and User Centric LIME. This unified approach addresses essential pitfallsin current XAI methodologies that have made the explanations better and more reliable, informative,and user-centric. The Feature Informative, Stable and User Centric LIME framework will set the wayforward for future work on XAI towards a more transparent and trustworthy representation of theinternal working mechanism of machine learning models for any application.