An LSTM-based system for accurate breast shape identification and personalized bra recommendations for young women

Abstract

Young women often struggle with ill-fitting bras, which can impede breast development and may contribute to breast-related health issues. From a business perspective, this also increases the likelihood of product returns. While various methods have been proposed to classify breast shapes and enhance bra design, the lack of accessible measurement tools and expertise often makes it difficult for young women to select appropriate bras. This study addresses this issue by introducing a Long Short-Term Memory (LSTM)-based recommendation system that helps young women easily and accurately choose well-fitting bras. To precisely identify the shapes of young women's breasts, this study collected human body data from 150 individuals using a 3D body scanner and employing ergonomics expertise and statistical analysis methods such as Principal Component Analysis (PCA) to classify 7 key indices such as bust circumference, underbust circumference, and lower mammary cup arc length as breast shape indices. The k-means clustering method subdivided breast shapes into six categories. Next, based on the classification result, an LSTM based discrimination model was developed to identify the breast morphology, achieving a breast identification accuracy of 85.71 %, surpassing both the Back Propagation (BP) neural network and Convolutional Neural Network (CNN) in terms of operational efficiency, fitting accuracy, and overall performance. Finally, users only need to input the breasts measurements to receive recommended bra parameters and corresponding style chart. This study not only enhances comfort and health by facilitating optimal bra selection but also minimizes return rates and improves the shopping experience, crucial for the sustainable development of the apparel industry.