GA-Based Optimization of Generative Adversarial Networks on Stock Price Prediction

Abstract

The combinational model of Generative Adversarial Networks (GANs) and Long Short-Term Memory (LSTM) has a good performance for analyzing and predicting time series data. The model performance, however, depends on the hyper-parameters which are defined in advance by the machine learning designers. In this study, Genetic Algorithm (GA) is applied for the design of hyper-parameters in the combinational model of GANs and LSTM in the stock price prediction. GA defines the candidate solutions as chromosomes. The GA chromosome includes the window size of training data and the hyper-parameters of the LSTM employed in the Generator of GANs. The experimental results show that the optimized model has a better performance than the original model.