Decision quality enhancement in Bayesian-like possibilistic classification of raw data using a new fusion strategy

Résumé

This thesis focuses on improving decision quality in fuzzy classification of imperfect data, which are mainly characterized by insufficiency and heterogeneity. To address these challenges, a new fuzzy decision-making approach based on the possibilistic framework is proposed for constructing membership functions. In classification problems involving low-quality data, ambiguous situations often arise when the belief degrees associated with competing classes are close to each other, leading to uncertain or unreliable decisions. To overcome this issue, we introduce a novel algorithm called G-Min (Generalized Minimum-based), which incorporates a reflexive decision mechanism that delays the final decision until a sufficiently reliable one is obtained. Based on this algorithm, two possibilistic fuzzy classifiers are developed. The first is designed for categorical data and extends Dubois et al.’s probability-to-possibility transformation to construct membership functions. The second is adapted to mixed data (categorical and numerical) and relies on a specific possibilistic approach for building membership functions for each data type. Experimental results on both synthetic and real-world datasets demonstrate that the proposed classifiers outperform several fuzzy and non-fuzzy benchmark methods for categorical and mixed data, respectively.

Citer ce document

Accès au document

Ce document est hébergé sur son dépôt institutionnel d'origine.

Auteur(s)

Statistiques

Consultations : 2

Téléchargements : 0