Deep learning-based prediction of women’s torso shapes from clothing pressure for tight-fitted top design

Purpose This study aims to develop a body shape prediction model by integrating three-dimensional (3D) digital clothing pressure simulation with artificial intelligence (AI), focusing on tight-fitting garment contexts. Design/methodology/approach The torso body shapes of women in their 20s and 30s were first categorized through factor analysis and cluster analysis based on direct anthropometric measurements. Representative 3D avatars were generated for each cluster, and fitted tops were virtually simulated using 3D clothing pressure analysis. The resulting pressure data were used to train a deep neural network (DNN) model for body shape prediction. In addition, unsupervised learning via t-SNE and K-means clustering was applied to compare classification outcomes. Findings The DNN-based model achieved an 88% prediction accuracy, effectively classifying body shapes based on clothing pressure distribution data. In contrast, the unsupervised clustering approach (K-means) produced a different classification structure, suggesting that pressure-based learning offers a distinct perspective from traditional anthropometric-based clustering. Originality/value This study is novel in employing clothing pressure distribution as an input for body shape prediction, rather than relying solely on conventional anthropometric data. It contributes a data-driven methodology applicable to garment design, virtual fitting and body-aware customization in 3D digital environments.