Weekend Box Office Forecasting Based on MF-DCCA and Multiple Neural Networks Approach

Abstract

The forecasting of box office receipts is always a difficult and challenging problem. In this paper, a forecasting approach is proposed to accurately forecasting weekend box office receipts based on multifractal detrend cross-correlation analysis (MF-DCCA) and multiple neural networks (DNN, CNN, LSTM). The MF-DCCA quantitatively finds that there is a long-range cross-correlation between each factor and box office receipts, which is multifractal. Its rolling window can reflect the intrinsic mechanism of neural network, and extract effective data that can improve the generalization ability of the model from the original data. The predictive performance of these frameworks is able to meet or exceed the method based on a single neural network. For (MF-DCCA)-DNN, the classification accuracy is 0.843, (MF-DCCA)-CNN is 0.786, and (MF-DCCA)-LSTM is 0.986. This research provides a theoretical method to further to enrich and improve the accuracy of box office receipts prediction and the extraction of effective feature.