Research On Clothing Sales Forecast Based On RBF Neural Network

Under the background of the epidemic,the number of users of tik tok e-commerce platform has grown rapidly in the past two years,and a large number of user traffic and the influx of brand merchants have contributed to the prosperity of the platform ecommerce ecology,after experiencing rapid growth in the number of brands and users,existing brands lost the dividend period of platform traffic growth and began to enter the stage of competitive operation optimization and resource integration.In order to improve the competitive advantage of the brand in the field of live broadcasting,accurate sales forecasting is particularly important,live broadcast sales forecast can help enterprises allocate inventory in advance,scientifically formulate marketing strategies,improve brand user satisfaction,enhance fan stickiness,and then improve brand profits.First,according to the data characteristics of different categories,determine the factors affecting live broadcast sales applicable to brand live broadcast rooms.The specific indicators collected 22 features such as number of viewers and peak number of people,and grouped these 22 features into six categories.By doing Spearman correlation analysis on the influencing factors and target sales,and applying Random Forest,Gradient Boosting Decision Tree(GBDT)and extreme gradient boosting(Xgboost)models to analyse the contribution of the features,the features with high correlation and contribution were screened out,and then collinearity diagnosis was performed on the features to eliminate the collinearity between the features.Finally,coupled with historical sales,the underwear category determined 6 input layer indicators,and the down jacket category determined 5 input layer indicators.The sales data of live broadcast are characterized by data clutter and large fluctuations.Through the research of existing sales prediction methods,RBF neural network has strong nonlinear fitting ability and is suitable for sales prediction of live broadcast.In view of the shortcoming that RBF neural network is easy to overfit,particle swarm optimization(PSO)and genetic algorithm(GA)are selected to optimize RBF neural network,and two optimization models of PSO-RBF and GA-RBF are constructed.Through the grey dolphin data platform,a total of 300 days of live broadcast data of the NEIWAI underwear flagship store from May 18,2021 to April 11,2022,and a total of 183 days of live broadcast data of Bosideng flagship store from October 1,2021 to April 1,2022,and using Python software to conduct modeling and simulation experiments.Experimental results show that compared with RBF neural network,the prediction effect of underwear category test set is as follows: MAE values of PSO-RBF and GARBF models decreased by 9.96% and 8.11%,MSE values decreased by 20.68% and4.86%,RMSE values decreased by 10.94% and 2.46%,and SMAPE values decreased by8% and 2%,respectively.Prediction effect of down jacket category test set: MAE value of PSO-RBF model and GA-RBF model decreased by 15.35% and 3.68%,MSE value decreased by 36.65% and 13.93%,RMSE value decreased by 20.41% and 7.23%,SMAPE value decreased by 10% and 7%,respectively.Through the comparison experiment,it can be seen that the optimization of RBF neural network by particle swarm optimization algorithm and genetic algorithm has better prediction effect on both seasonal and non-seasonal category data sets.Particle swarm optimization algorithm is more suitable than genetic algorithm to optimize RBF neural network for the prediction of live broadcast sales.Finally,according to the research conclusion,the shortcomings of the paper are summarized and make corresponding suggestions for enterprises.