基于Stacking模型融合的店铺销量预测

Store sales forecast based on Stacking model fusion

为了实现企业产品销量预估,提高生产供应的准确性与效率,提出了基于Stacking模型的融合算法进行销量预测。算法设计了两层堆叠的模型结构,初级学习器采用随机森林、支持向量回归、差分整合移动平均自回归、轻量级梯度提升机器和门控循环单元5种单模型,将分类与回归树作为次级学习器构成Stacking融合模型,并对数据进行了预测。预测结果显示,使用Stacking模型融合后得到了较好的预测结果,比单模型中效果最好的模型的均方根误差更小,平均绝对误差更小,决定系数值更大,表明Stacking融合模型的预测准确率更高。所设计模型可用于对企业店铺的产品销量进行预测,帮助企业更好地安排生产、营销活动,为减少库存、缩短生产销售周期提供数据支持,对企业生产决策有一定的参考价值。

In order to realize the sales volume prediction in modern enterprises and achieve the accuracy and efficiency of production and supply,a fusion algorithm based on Stacking model was proposed for sales volume prediction.The algorithm designed a two-layer stacked model structure.The primary learner adopted five single models:RF(random forest),SVR(support vector regression),ARIMA(autoregressive integrated moving average),LGBM(light gradient boosting machine),GRU(gated recurrent unit),CART(classification and regression tree)was used as secondary learners to form a Stacking fusion model,and the data were predicted.The prediction results show that there are better prediction results in the relevant data after using the Stacking model fusion.Compared with the model with the best effect in the single model,the RMSE(root mean square error)is smaller,the MAE(mean absolute error)is smaller,and the R2(coefficient of determination)value is larger,indicating that the prediction accuracy of the Stacking fusion model is higher.This model can be used to predict the product sales and other relevant data of enterprise stores,help enterprises better arrange production and marketing,provide reference data for reducing inventory and shortening production and sales cycle,and have a certain reference value for enterprise production decision-making.