Dance classification using pretrained deep learning models integrated with the circular Fermatean fuzzy MARCOS method

Abstract

Pretrained deep learning models offer a method for classifying dance by analyzing movement behaviors in videos or using sensors to assess the degree of automation in recognizing different types of dances. These models utilize transfer learning strategies to enhance recognition accuracy across various datasets. This study presents a hybrid framework for identifying and classifying dance styles using pretrained deep learning models, evaluated under multiple performance criteria. To address the inherent uncertainty and complexity in selecting the best-performing models, a novel circular Fermatean fuzzy measurement of alternatives and ranking based on the compromise solution (CFF-MARCOS) approach is proposed. Unlike existing methods, the integration of circular Fermatean fuzzy sets (CFFS) into the MARCOS framework enables more refined handling of hesitation and ambiguity in expert evaluations. A case study involving ten pretrained models, seven criteria, and three experts demonstrates the superiority of the proposed method in generating robust and interpretable rankings. Results highlight improved decision reliability and clarity in model selection for automated dance classification tasks.