摘 要：神经网络被广泛应用于目标检测、优化组合等领域，但其往往容易过拟合。为解决过拟合问题，通常对神经网络稀疏化，这类技术目前较为成熟，如 dropout。文章主要考虑在 Lasso 罚函数情形下，通过对神经网络连接的权重进行压缩，实现高维非线性情形下的变量选择，并使用蒙特卡洛模拟验证该稀疏神经网络的变量选择结果具有一致性。最后将该模型应用到重庆市火锅团购销量分析中，得到 10 个对火锅销量最具影响的因素。
关键词：神经网络；稀疏神经网络；变量选择
DOI:10.19850/j.cnki.2096-4706.2022.06.021
中图分类号：TP399 文献标识码：A 文章编号：2096-4706（2022）06-0086-04
Analysis of Influencing Factors of Hot Pot Sales Based on Sparse Neural Network
GUO Ping
(School of Mathematics and Statistics, Guangxi Normal University, Guilin 541006, China)
Abstract: Neural network is widely used in the field such as target detection, optimization and combination and so on. but it is easy to overfit. In order to solve the overfitting problem, neural networks are usually thinned and such techniques are mature, such as dropout. This paper mainly considers the variable selection in the high-dimensional nonlinear case by squeezing the weight of the neural network connection under the Lasso penalty function case. Monte Carlo simulations are also used to verify the consistency of the variable selection results for this sparse neural network. Finally, the model is applied to the sales analysis of Chongqing hot pot group purchase, and 10 factors that have the most influence on the sales of hot pot are obtained.
Keywords: neural network; sparse neural network; variable selection
参考文献：
[1] 周书豪 . 神经网络在基因型不确定数据和经济数据上的研究 [D]. 桂林：广西师范大学，2020.
[2] KRIZHEVSKY A，SUTSKEVER I，HINTON G E. ImageNet Classification with Deep Convolutional Neural NetworksBy Alex Krizhevsky [C]//Advances in neural information processing systems.2012：1097-1105.
[3] LIANG F，LI Q，ZHOU L. Bayesian Neural Networks for Selection of Drug Sensitive Genes [J].Journal of the American Statistical Association，2018，113（523）：955-972.
[4] GIRSHICK R，DONAHUE J，DARRELL T，et al. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation [J].IEEE Computer Society，2013.
[5] FENG J，SIMON N. Sparse-Input Neural Networks for Highdimensional Nonpara-metric Regression and Classification [J].Statistics，2017.
[6] SUN Y，SONG Q F，LIANG F M. Consistent Sparse Deep Learning：Theory and Computation [J].Journal of the American Statistical Association，2021，1-42.https://doi.org/10.48550/arXiv.2102.13229.
[7] 周徐达 . 稀疏神经网络和稀疏神经网络加速器的研究 [D].合肥：中国科学技术大学，2019.
[8] 美团，中国连锁经营协会 .2021 中国餐饮加盟行业白皮书[EB/OL].[2021-11-08].http://www.ccfa.org.cn/portal/cn/xiangxi.jsp?id=442768&type=33.
[9] CY373.2020 年中国火锅餐饮行业分析：规模不断增长、两级分化严重 [EB/OL].[2021-11-08]. https://www.chyxx. com/industry/202108/966439.html.
[10] CY331.2021 年中国火锅产业规模及龙头企业对比分析：海底捞 VS 呷哺呷哺 [EB/OL].[2021-11-08].https://www. chyxx.com/industry/202109/975214.html.
[11] ROBERT，TIBSHIRANI. Regression Shrinkage and Selection via the Lasso [J].Journal of the Royal Statistical Society. Series B：Methodological，1996，58（1）：267-288.
[12] MAO Y，YAN S. Variable selection via penalized neural network：a drop-out-one loss approach [C]//in International Conference on Machine Learning，PMLR，2018：5620-5629.
作者简介：郭萍（1998—），女，汉族，广西钦州人，硕士研究生在读，研究方向：数理统计。