Purpose The apparel and fashion industry is one of the most dynamic industries, constantly changing in response to market trends and increasing consumer demand for product quality. In this sector, garment fit criteria greatly influence product development and fashion quality control. This study aims to propose a decision-support framework by integrating standardized fabric mechanical properties with 3D virtual prototyping and machine learning (ML) models to predict deformation-related indicators of garment–body interaction, including deformation region, strain and stress, to support early-stage pattern development, material screening and more efficient virtual prototyping. Design/methodology/approach The research process consisted of four stages: fabric mechanical data collection, and data set construction; data preprocessing and feature selection to ensure data accuracy; developing a predictive model using a sequential decision tree–linear regression approach and a multi-target random forest (MT-RF); and evaluation and verification of the model’s performance in early-stage pattern development. Findings The sequential model generated clear decision rules and region-specific regression equations, achieving a classification accuracy of 78.3%. The MT-RF achieved 97.0% accuracy in deformation-region classification. Notably, the coefficient of determination was 0.916 for strain and 0.884 for stress, confirming robust joint prediction of categorical and continuous outcomes. The predicted deformation indicators enabled the identification of localized deformation-prone zones, providing useful guidance for early-stage pattern refinement and fabric selection in digital garment development. Originality/value This study contributes a design-oriented decision-support framework by translating standardized fabric measurements into deformation-related indicators for decision-making in garment development. It offers an interpretable and application-oriented framework that supports early-stage pattern refinement, material screening and more efficient virtual prototyping.
Phan et al. (Sat,) studied this question.