Classification of Clothing Products Based on Principal Component Analysis and Clustering Algorithm

Abstract

Inventory management of clothing products becomes more and more difficult due to its high volatility and short life cycle features. In order to predict the stock condition of each stock keeping unit, this article presents a condition forecasting method based on principal component analysis and weighted K-means algorithm, which can classify the sales condition before the following sales period. Firstly, two common clustering algorithms and three supervised learning algorithms are tested by real world data of a Chinese clothing company. The results indicate that the supervised algorithms perform better, which illustrates potential correlation between adjacent selling periods. Secondly, the correlate factors and attribute weight are estimated using principal component analysis (PCA). After that, a group of multi-factor weighted K-means algorithms are proposed and tested on the data set. The results show that a three-factor weighed algorithm has better performance than the others, which can make full use of the correlation factors between adjacent sales periods, therefore promises an 80% overall accurate classification rate.