This study aims to train supervised machine learning algorithms on 3D body-scanned anthropometric data and develop an automated model for accurately predicting female body shapes.
A sample of 211 adult females participated in this study to identify body shapes and the key body measurement associated with them. The SizeUSA database was imported to train and validate the predictive models based on six supervised machine learning algorithms: multinomial logistic regression (MLR), random forest, support vector machine, naïve Bayes, decision tree and artificial neural network. The models were compared using performance evaluation metrics such as accuracy, precision, recall and F1-score. MLR demonstrated the highest accuracy and was used to classify participants into three distinct body shape groups. The MLR-based model classified the 211 participants into three distinct body shape groups and identified key body dimensions that significantly influenced classification likelihood for each group. For additional validations, these key measurements were compared across three body shape groups using analysis of variance.
The MLR-based body shape predictive model captured more comprehensive and nuanced anthropometric relationships than the Female Figure Identification Technique (FFIT) formula, a widely used method in apparel research for body shape classification. In addition, a comparison of the three body shape groups demonstrated that the MLR-based predictive model effectively captured the unique characteristics of each group, offering a more precise and detailed classification approach.
This study highlights the effectiveness of developing a novel predictive model that can rapidly identify body shapes from large data sets while identifying additional dimensions beyond the standard measurements used in the FFIT formula and sizing system.
