Study On The Mid-term Sales Forecasting Of Clothing Based On The Theory Of Extreme Learning Machine

With the textile and clothing industry continues to develop and mature, brand developmenthas become a mainstream trend in this industry which occupies very important status within thepeople’s livelihood oriented industries in China. At present, the domestic brand clothingoccupying certain market share mainly conduct its sales through the establishment of franchiseemodel. Aiming at the completion of preparation task for the commodity, the franchisees in theorder fair need to place order of the clothing to be sold in the future, thus requiring them toforecast future sales and make the purchasing plan as well as the procurement budget based onthe historical sales data. However, The franchisees’ operating difficulty was increased due to theaccumulating of large volume of inventory in the end of sales season that has to be assumed bythemselves solely as a result of the unreasonable procurement decision based on the predictedvalue that cannot reflect the future sales situation because of the inaccurate sales forecastingresult for franchisees at present in the franchise model of “buyout” mostly adopted by the brandclothing. The sales forecasting for the entire quarter belongs to the category of medium-termsales forecast as the franchisees’ order is geared to the whole season.The extreme learning machine system of artificial neural network is selected as the finalprediction method in this paper as the characteristic of the problem to be solved is in conformitywith the ability of extreme learning machine neural network of artificial neural network to utilizethe historical data of the same period in previous years to predict the sales amount of theidentical period in current year and avoid the defection of artificial neural network being slow inthe learning speed and prone to fall into local minimum value in the meantime in addition to thediscovery of artificial neural network of machine learning theory being capable of mapping anycomplex linear and nonlinear relationship and having strong ability of learning andgeneralization through the study of a large number of prediction methods making it very suitablefor solving the problem of clothing product sales forecasting.In determining the prediction method, this article first complete the construction of topologystructure of network and the settlement of network coefficients,by building a model which have4input layer number of nodes,6hidden layer number of nodes,1output layer number of nodesand its activation function is S function. An empirical test is proceeded in this paper followed bythe completion of the construction of network topology structure and the settlement of network coefficients. The test detected the extreme learning machine has a poor stability. Thecomparison of the forecast model been optimized with the one before optimization reveals ahigher stability of the former with the extreme learning machine neural networks beingoptimized by introducing the integration theory.To further verify the prediction performance of the optimized model, its prediction resultis in comparison with the actual value and the prediction result of BP neural networkrespectively, both comparison approaches adopted for the test validating that the extremelearning machine neural network optimized by the theory of integration has superior advantageand better reliability.Results of the study in this paper indicate that the ensemble prediction model of extremelearning machine has better predictive effect in the clothing sales forecasting enough to providetheoretical support for the franchisee’s procurement decision.