Multiple regression prediction combined with deep learning is a very effective method—compared with traditional machine learning models, its prediction ability is better. Various neural networks perform very well in multiple regression prediction and can learn a series of advanced feature representations in the data, so as to analyze the complex relationship between the data and improve work efficiency. The application of long short-term memory (LSTM) neural network to multiple regression prediction has great advantages. It can not only solve some ordinary prediction problems but also deal with temporal prediction problems or some regression prediction with complex relations with multiple features. Hyperparameter adjustment of LSTM neural network will directly affect the accuracy of multiple regression prediction. How to effectively find out the appropriate hyperparameter is one of the problems in the field of neural network. In this paper, a new metaheuristic algorithm bamboo forest growth optimization (BFGO) algorithm is used to optimize the hyperparameters of LSTM neural networks. Through the efficient exploit and exploration ability of BFGO algorithm, the hyperparameters which are most suitable for LSTM neural network to solve the prediction problem are found. We selected two datasets related to multiple regression prediction to test the optimized model, and the results show that the accuracy of the optimized LSTM model using BFGO algorithm is better than that of other models.