LSTM Neural Network with Parallel Swarm Optimization Algorithm for Multiple Regression Prediction

Abstract

Multiple regression prediction can be applied to many applications, including predicting market prices, weather conditions, and so on. In recent years, the use of neural network models to deal with some problems related to multiple regression prediction has been proposed. The accuracy of prediction with long short-term memory (LSTM) neural network will be affected by the model parameters. If the parameters are adjusted according to human experience, there will be some limitations. In this paper, we propose a method to optimize multiple regression prediction of LSTM neural networks with a gannet optimization algorithm (GOA) modified by parallel communication strategies. The method optimizes the three parameters of the count of nodes in the hidden layer, training epochs, and learning rate of the LSTM neural network by the parallel gannet optimization algorithm (PGOA) to improve the accuracy and reliability of the prediction. The data results from the 28 benchmark functions tested by CEC2013 show that PGOA is more capable of finding the optimal solution compared to other algorithms. The accuracy of the PGOA-LSTM model and other models is tested with two datasets. The experimental results show that the PGOA-LSTM model predicts the data with higher accuracy than other models.