Research On Refrigerator Sales Forecast Based On Web Search Data

With the continuous improvement of people's living standards,people's demand for refrigerators has gradually changed,and the refrigerator industry has also initiated a new generation of intelligent reforms.Therefore,it is reasonable and important to predict market demand and make decisions on manufacturers'material,production planning,market planning,etc.However,there are few studies on refrigerator sales forecasting at present.Most of the only studies use traditional gray network or time series models for prediction.The feature selection is relatively broad,and the accuracy of the prediction results is not high.It isn't helpful for a production plan for modern enterprises.With the increasing maturity of the Internet era,search engines are closely linked to everyone's lives,and prediction research based on the search index has been applied in many fields and achieved good results.In addition,the popular of online shopping malls has made online shopping becomes the mainstream consumption method and the large number of online reviews generated by online shopping have become one of the important basis for people to purchase goods.In summary,this article takes refrigerators as the research object,based on the web search data including Baidu index and sentiment index,use four feature selection methods to select the feature subsets that have the best impact on refrigerator sales,and build a variety of machine learning algorithms to predict monthly refrigerator sales.The research content of this article mainly includes the following aspects:?1?Analyze consumers'decision-making behavior when purchasing refrigerators based on consumer purchase decision theory,propose web search data including Baidu index and sentiment index,and construct a correlation framework model of web search data and refrigerator sales,theoretically analyze the ability of web search data to predict refrigerator sales;?2?Obtain the Baidu index and sentiment index in the web search data and verify the correlation between two index with the refrigerator sales,which lays the data foundation for the prediction model later;?3?Based on the traditional Wrapper heuristic method and the adaptive genetic algorithm proposed in this paper,different machine learning algorithms are used as the base model for feature selection of the web search data,and the feature subset with the best impact on the refrigerator sales forecast is selected for model establishment;?4?Based on the traditional time series model and two machine learning algorithms of random forest and support vector machine,combined with four feature subsets,at the same time,based on whether the feature subset contains sentiment index as a comparison,a total of 17sets of refrigerator sales prediction models were established,and analysis different model's prediction result.The results show that the feature subset obtained by adaptive genetic algorithm using random forest as the base model for feature selection has the best performance,and the evaluation of various indicators of the prediction model based on random forest is better.The minimum error rate MAPE is 2.33%,increased 2.87%compared with the best results 5.2%of previous studies^[73].