Izvorni znanstveni članak
https://doi.org/10.24138/jcomss-2025-0070
P-LIME: PSO-based Local Interpretable Model-Agnostic Explanations Approach for More Reliable AI Explanations
Khaled Bechoua
; Constantine 2 University – Abdelhamid Mehri, Algeria
*
Ahmed Taki Eddine DIB
; Constantine 2 University – Abdelhamid Mehri, Algeria
Hichem Haouassi
; Abbas Laghrour University, Khenchela, Algeria
Hichem Rahab
; Abbas Laghrour University, Khenchela, Algeria
* Dopisni autor.
Sažetak
The rapidly expanding fields of artificial intelligence and machine learning see growing use of intelligent models for predictive tasks and decision support in areas like healthcare, autonomous transportation, and finance. However, the absence of transparency in these models makes them difficult for end-users to understand, limiting their trust and adoption. To address this challenge, techniques such as Local Interpretable Model-agnostic Explanations (LIME) have been developed to provide local model-agnostic explanations independently of the model’s internal structure. Despite its effectiveness, LIME suffers from limitations in the random generation of perturbed instances, which can lead to unstable and low-quality explanations. To handle these drawbacks, this work introduces a PSO-based Local Interpretable Model-Agnostic Explanations (P-LIME) approach. P-LIME leverages the Particle Swarm Optimization (PSO) metaheuristic to intelligently generate perturbed instances, thereby improving the quality and stability of the explanations. Experimental results demonstrate that the proposed approach significantly outperforms the original LIME method, offering a more reliable and interpretable framework for understanding complex artificial intelligence models. This advancement contributes to the broader goal of enhancing transparency and trust in artificial intelligence systems.
Ključne riječi
Explainable artificial intelligence; interpretable machine learning; Local Interpretable Model-agnostic Explanations (LIME); Particle Swarm Optimization; Model-agnostic Explanations; metaheuristics; Trust in artificial intelligence
Hrčak ID:
336742
URI
Datum izdavanja:
15.7.2025.
Posjeta: 73 *