Kernel-based clustering approach in developing apparel size charts

Abstract

With the industry revolution, apparel products also become more sophisticated moving from the basic purpose of clothing to aesthetic appeal of the garment embracing the concepts garment fitting and fashion. Garment fitting is a key technical essential for comfortable wearing. In garment fitting, size refers to a set of specified values of body measurements, such that it will provide a means for garments perfectly fit to a person. With the advent of computer software and improved data mining techniques, researchers attempted new advances in formulation of size charts with a better fit. This article suggests a kernel-based clustering approach in developing an effective size chart for the pants of Sri Lankan females. A new kernel based approach “Global Kernel K- means clustering” was successfully deployed to cluster lower body anthropometric data of Sri Lankan females within the age range of 20-40 years. Through the proposed Kernel- based clustering method can effectively handle highly non-linear data in input space which is a key property of lower body anthropometric data and make it linearly separable in feature space without reduction in dimensions and also mathematically justified. Through this method promising results could be obtained and further clustering method was internally validated with kernel based Dunn’s index. The level of fitness of the developed size chart was also evaluated with the aggregate loss of fit factor. The proposed method has strong implications to utilize globally in developing size charts.