Research On Factors Screening Based On Random Forest And Domestic Movie Box-Office Combined Prediction Model

With the rapid development of the film industry,more and more investors are keen on film investment,but at present,only a small number of film investments in China’s domestic film investment are profitable.In this context,the prediction of domestic movie box office will undoubtedly have great practical significance for risk control.As a typical short-cycle and experiential product,the movie box office revenue is affected by many factors,such as the characteristics of the film itself and market-related factors.With the popularity of the Internet,network information has also become an important factor affecting users’ decision-making.Therefore,how to build a more comprehensive movie box office influence factor indicator system,how to better use the network information to predict domestic movie box-office is a focus of current research.Based on the summary of the research status and the actual situation of the industry,this paper constructs a new domestic box office prediction model to improve the shortcomings in the current research and improve the accuracy of the box office prediction model:(1)considering the current domestic box office prediction,most of the model research has the problem of insufficient data dimension.This paper builds a comprehensive set of domestic box office influence factors from a number of dimensions and combines the status quo of China’s film industry,and makes full use of the quantitative methods of influencing factors.The network platform data makes the quantitative value of the influencing factors more time-efficient;(2)considering the excessive input variables will increase the difficulty of model training,so this paper builds a Random Forest model to calculate the influence of factors,and use this as a basis to screen the influencing factors,to simplify the input of subsequent predictive models;(3)Because the influence of different influencing factors on the box office are different,in order to solve the problem of the average weight distribution of the influencing factors in the previous research,this paper constructs a box office prediction model based on weighted K-Means and local BPNN: The weight of each factor is determined by factor influence,the samples are divided by weighted K-Means clustering,and local BPNN model was constructed based on subsamples for box office prediction.Finally,in order to verify the superiority of the model constructed in this paper,415 domestic film related data released in 2016-2017 are collected,and the quantitative methods and models defined in this paper are used to conduct experiments,and the domestic movie box office prediction model based on BPNN and Based on K-Means and local BPNN,the domestic box office prediction model is compared.The experimental results show that the Mean Absolute Percentage Error(MAPE)of the model constructed in this paper is 8.62%,which is lower than the domestic experiment based on BPNN box office prediction model(13.91%)and the domestic movie box office prediction model based on K-Means and local BPNN(11.16%);the Root Mean Square Error(RMSE)is 238.26,which is lower than the comparison experiment based on BPNN Model(516.73)and domestic movie box office prediction model based on K-Means and local BPNN(367.41).It can be seen that the domestic box office prediction model constructed in this paper can improve the prediction accuracy of the box office to a certain extent.