基于线性混合模型的电商销售量的预测研究

【作者】 王宁；

【导师】 王艺舒；

【作者基本信息】 青岛大学 ， 应用统计（专业学位）， 2019， 硕士

【摘要】 21世纪以来,随着科技和互联网的飞速发展,互联网企业获得了广泛的发展空间,各电商平台也在一夜之间崛起,而一个电商企业的良好发展离不开对自己产品走势的良好把握,古者云,“凡事预则立,不预则废”。但是,销售量的预测问题会受很多方面因素的影响,而且还有很多不确定性的影响因素会导致销售量出现变化,虽然以往的时间序列分析方法已比较成熟,而且应用也比较广泛,但每种方法也有其自身的局限性,如ARIMA模型的定阶问题、神经网络模型复杂程度的问题、支持向量机核函数的选择问题等,往往会导致预测的效果达不到人们的预期水平。本文提出一种利用线性混合模型(LMM)来对销售量进行预测的建模方法,结合特征工程的相关知识,进行特征构造和特征选择,并根据已经选取的特征,进一步选择模型的固定效应与随机效应,通过选取的随机效应以反映时间序列的季节性、周期性以及其他不确定性因素的影响等问题,有别于传统的时间序列分析模型。利用京东商城近四年来的销售量数据进行实证分析,并且对比分析神经网络和支持向量机模型的预测效果,说明本方法的有效性。以帮助企业决策层对未来销售量的走势有一个整体把握,从而对资源可以进行有效的配置,合理布置发展规划,以提高企业在同行中的竞争性。更多还原

【Abstract】 Since the 21st century,with the rapid development of technology and the Internet,Internet companies have gained a wide range of development space,and e-commerce platforms have also emerged overnight.The good development of an e-commerce company is inseparable from the good grasp of the trend of its own products.The ancients cloud,"everything is pre-established,not pre-emptive." However,the forecast of sales volume will be affected by many factors,and there are many uncertain factors that will lead to changes in sales.Although the previous time series analysis methods are mature and widely used,each method has its own limitations,such as the fixed order problem of ARIMA model,the complexity of neural network model,and the support vector machine kernel function.Choosing problems,etc.,often leads to the prediction that the effect is not as high as people expected.This paper proposes a modeling method that uses the mixed linear model(LMM)to predict sales volume,combines the relevant knowledge of feature engineering,performs feature construction and feature selection,and further selects the fixed effect of the model based on the selected features.Random effects,different from traditional time series analysis models,by choosing random effects to reflect the seasonality,periodicity,and other uncertainties of time series.The empirical analysis of the sales volume data of Jingdong Mall in the past four years is carried out,and the prediction effects of neural network and support vector machine model are compared and analyzed,and the effectiveness of the method is illustrated.To help the decision-making level of the enterprise to carry out the layout of production resource planning and market development and development,further deepen the development of the enterprise and enhance the competitiveness of the enterprise.更多还原