Exploring aesthetic judgments of modern sculptures in virtual reality: a multimodal machine learning approach

Abstract

Most research on art perception has focused on two-dimensional works, leaving three-dimensional media relatively underexplored. This study presents an exploratory multimodal approach to investigate how non-expert viewers perceive and evaluate modern sculptures in a custom-made virtual reality (VR) environment. Twenty-eight participants without formal art training viewed 3D scans of 10 figurative sculptures across three rounds. Participants first provided aesthetic and emotional judgments, then evaluated specific sculptural attributes, and finally repeated the aesthetic and emotional ratings. Eye tracking, pupil responses, and electroencephalography (EEG) were recorded throughout. Gradient Boosting Decision Trees (GBDT) combined with SHAP (SHapley Additive exPlanations) analysis were used to examine relationships between sculptural attributes and subjective evaluations. Beauty, Valence, and Technical Skill were the judgments most consistently related to rated sculptural attributes, with Visual Harmony and Texture emerging as the most influential predictors. Following exposure to attribute descriptions, aesthetic evaluations became more systematic, reflected by an increase in explained variance for Valence from 68% to 88%. Physiological analyses further revealed differences in visual engagement and EEG signal characteristics between participants with contrasting aesthetic evaluations. Participants who rated sculptures as more beautiful tended to exhibit fewer fixations and saccades, longer fixation durations, and lower gamma-band power. The findings demonstrate the feasibility of combining VR, interpretable machine learning, and multimodal physiological measures to investigate three-dimensional aesthetic experience. The observed relationships between perceptual attributes, subjective judgments, and physiological responses provide preliminary descriptive evidence of potentially relevant patterns in sculpture perception. However, given the use of single-channel in-ear EEG, the absence of ocular artifact correction, and data-quality constraints, the physiological findings should be interpreted as preliminary correlations rather than evidence of specific neural mechanisms. Larger samples and more rigorous neurophysiological methods are needed to validate these findings and further characterize the mechanisms underlying three-dimensional aesthetic experience.