Abstract To address the issues of lacking scientific references for film investment decisions and low prediction accuracy of single models, this paper integrates the extreme gradient boosting (XGBoost) algorithm with a fully connected neural network (FCNN) to construct the XGBoost-FCNN model. First, the Bayesian optimization algorithm is used to tune the hyperparameters of XGBoost. A two-stage optimization strategy combining particle swarm optimization and Bayesian optimization is employed to optimize the hyperparameters of FCNN. Then, the sequence least squares planning algorithm is used to minimize the error between the prediction values of the two models, thereby determining the model fusion weights. Finally, experimental results based on 7, 668 publicly available IMDb data points from the Kaggle platform demonstrate that the model achieved an R2 value of 0. 7240 on the test set, with MAE of 0. 7048, MAPE of 4. 5858, MSE of 1. 0678, and RMSE of 1. 0334. This indicates that the experimental model outperforms XGBoost, FCNN, and gradientboost models in terms of stability and accuracy, demonstrating a certain degree of resistance to overfitting and providing a feasible solution for movie box office prediction.
Wu et al. (Tue,) studied this question.