As a typical short cycle and experiential product, Movie's box-office is influenced by many factors, so it is hard to forecast its box-office accurately. In this study, a box-office forecasting model based on weighted K-means and local BP Neural Network (BPNN) is constructed, with aims to improve the shortcomings of the current model and improve the accuracy of box office prediction:(1) Construct the factor influence measurement model based on Random Forest (RF) and simplify the box-office influence factors according to the value of variable importance, to achieve the purpose of simplifying the input of the following forecasting model. (2) In the traditional researches, the weight of each factor was equally allocated in sample classification, which without considering the question of different factor has different influence. So a box-office forecasting model based on weighted K-means and local BPNN is constructed, using weighted K-means clustering to classify the samples based on the value of factor influence, then build several local BPNN models based on each subsample. Experiments show that the Mean Absolute Percentage Error (MAPE) of this study's model is 8.49%, which is lower than 10.39% of the contrast experiment, which proves the superiority of the box-office forecasting model built in this study.