Fuzzy Decision Support for Sustainable Dance Heritage Preservation Using AI-Driven Narrative Modelling

Abstract

The preservation of intangible cultural heritage (ICH) is becoming progressively more difficult due to environmental disruptions and the intricate nature of recording historical performance traditions. Conventional preservation approaches remain constrained by the absence of scalable mechanisms capable of ensuring reliable decision-making within uncertain environments. This study introduces a fuzzy-based decision support architecture integrated with Artificial Intelligence (AI)-oriented narrative modelling to facilitate the sustainable preservation of dance heritage. Initially, a systematically organised representation of Chinese Lion Dance movements is developed to capture both the foundational motion patterns and the expressive attributes embedded within this traditional performance practice. The proposed framework is subsequently connected to sensor-generated data streams, thereby establishing a computationally executable platform for automated motion acquisition and transcription. The Dance Preservation Dataset (400k), obtained from Kaggle, undergoes pre-processing through Z-score normalisation, after which Principal Component Analysis (PCA) is employed for feature extraction to minimise dimensional complexity while preserving the most influential movement-related attributes. The recorded performances are then transformed into Labanotation, enabling structured archival documentation and facilitating the synthesis of new performances without compromising the authenticity and structural integrity of traditional dance expressions. To accomplish this objective, a hybrid neuro-fuzzy model is designed by integrating Sugeno-type fuzzy-mutated Enhanced Recurrent Neural Networks (STFuzzy-ERNN) with Sugeno-type fuzzy inference systems (SFIS). The STFuzzy-ERNN component performs movement pattern identification, whereas the SFIS module delivers interpretable and transparent decision-making capabilities.  The proposed framework assesses several preservation-oriented criteria, including movement intricacy, expressive richness, and historical relevance, thereby generating explainable recommendations for conservation prioritisation and archival management. Experimental outcomes demonstrate superior performance, achieving precision of 0.9892, recall of 0.9875, inference time of 6.4, and model size of 47.3. Furthermore, feature-importance evaluation identifies the most influential movement characteristics and sensor modalities contributing to preservation prioritisation. The findings establish an extensible and data-centric strategy for protecting intangible dance heritage. Through the integration of AI, fuzzy reasoning, and narrative modelling, the framework enhances long-term documentation, analytical interpretation, and intergenerational transmission of traditional dance practices for educators, performers, and cultural heritage organisations.