Research And Application Of Box Office Prediction Based On Fuzzy Linear Regression

With the improvement of film technology and the accelerated operation of the market,more and more film investors invade film industry,which also contributed to the rapid growth of Chinese film box office.But the film production is an expensive and adventurous attempt.Throughout the distribution of the entire box office,a wired phenomenon is that a small amount of high box office film pulls the growth of the entire movie box office and soars box office revenue to cover up the vast majority of the fact that the loss of the film.Therefore there’s an urgent demand for an effective film investment decision-making system to assist investors to ensure that the rate of return on investment.In the early stages of filming,the forecast is more difficult,as a result of the ill data itself and the larger difficult of prediction,the same director may also shoot "bad film",in the similar way,for actors,too.So it’s extremely hard to forecast the film pre-box office,plus the related research is rare,and most normal research is about the film production before the completion of the box office forecast,or the film released a week after the box office forecast.Although the use of the first box office data to predict the highest rate of film box office reaches up to 90%,there’s not much reference value for the early investors.In this paper,a method based on fuzzy linear regression is presented,which uses the basic information such as director,actor,type and distribution area before film shooting,and the historical box office data from 2011 to 2015 to predict the box office in 2016.In the process of quantifying the basic information of a film,it’s essential to take the director and actor’s influence fading with time into account,then make full use of directors and actors involved in all the cumulative film box office that influence,and manage the corresponding normalized quantification.Then adapt the EM and Kmeans clustering to cluster analysis(that is performing fuzzy linear regression analysis in different category.The large error in forecasting box office leads to that the average relative percentage error can not reflect the real forecast situation perfectly.Therefore,this paper presents an evaluation index based on the probability distribution function of relative percentage error,which compare with the average relative percentage error,and combine with the performance evaluation of the algorithm at the same time.In this paper,we use the triggering degree and the normal fuzzy number to predict the ambiguity minimization.We find that the predicted range is the same.When the confidence level h is the same,the interval and fuzziness obtained by using the normal fuzzy number is smaller.In the case of using the normal fuzzy number,the fuzzy effect minimization method is compared with the least square method.In the experimental comparison,it is found that when the normal fuzzy number is solved by the least squares method,the proportion of the sample with less than 50% is higher,the overall prediction error is smaller and the fitting degree is higher.On considering the impact of inflation,the adjustment of the weight model will improve the forecast.To sum up,the fuzzy linear regression algorithm and BP neural network algorithm are the least error in predicting the first week box office while the error is the biggest at the third week.The method of fuzzy linear regression using two different fuzzy numbers has better stability,faster speed and lower error than BP neural network.Therefore,the box office prediction model based on fuzzy linear regression has better practical significance.