Abstract: How to achieve fast and precise size-archiving in large numbers is a key issue to constructing the digital mass customization system of garments. Combining with sizes of the designed pivotal parts of women pants，this paper expanded the control parts of bottoms into 10 items based on the national standard; Taken female trousers as an example, after constructing simple size classification model with BP neutral network, the five control parts that have a major impact on the result of archiving could be screened through Mean Impact Value (MIV), and they were then conducted as the input layer of ensemble clothes size classification model. In the model, Height, waist circumference, hip circumference, waist height and back gore length worked as input variables of height model and waist circumference, hip circumference, thigh circumference, waist length , height were waist circumference model input variables. Then, after integrating 10 simple BP neural networks through Adaboost algorithm, it achieved the size-archiving model with high precision and strong generalization ability.

Key words: mass customization, MIV, size grading, neural networks ensemble