Forecast Of Brand Car Sales Based On Web Search Data

The automobile industry is the representative of the national industrial system,and the size of the automobile market directly affects the economic development of the country.Accurately grasping the forecast of automobile sales is of great significance to management departments and automobile manufacturers.This article predicts brand car sales based on online search data.First introduce consumer decision theory,and establish a theoretical framework of the relationship between online search and car sales;based on sales data to establish a time series forecasting model as a reference,analyze the deficiencies in the sales time series forecasting model;construct search keyword library based on established relationship theory framework,build different feature sets according to feature selection method;build machine learning prediction model,predict different feature sets and compare forecast effects.The main research contents are as follows:(1)Build a model framework of the relationship between online search and car sales.Introduce consumer purchase decision theory,explore the relationship between online search and car sales from the perspective of consumer purchase decision,and establish a theoretical framework of relationship.(2)Obtain the data of automobile sales and establish the time series model.First,the serial data were split to check the seasonality.The SARIMA model was established and predicted.Based on the prediction results,this paper discusses the shortcomings of the established sales time series prediction model and further introduces the network search data to predict.(3)To construct a database of search keywords and a subset of features.Based on the established relational model framework,the core search keywords are determined.Through the web crawler technology to obtain and collate the index data of search keywords.Based on a variety of feature selection methods,three feature subsets based on different selection methods are constructed.(4)Based on Random Forest and Gradient Boosting Tree,a prediction model is established and predicted based on the established feature subsets.The predict results show that the established machine learning model has excellent prediction effect.The MAPE of the GBDT prediction model based on the RFE method is 3.39%.Finally,the article gives some opinions on the trend of Volkswagen brand car sales and puts forward some countermeasures.