A Stock Prediction Method Based on Singular Spectrum Analysis and Long Short-Term Memory Network

There are lots of noise influence in stock price time series. Therefore, A stock prediction method based on Singular Spectrum Analysis (SSA) and Long Short-Term Memory network (LSTM) is proposed. SSA is used for stock time series denoising since it is appropriate to the nonlinear data denoising. The stock price prediction is realized using LSTM since it can effectively solve the problem of gradient explosion existing in traditional neural networks. To highlight the performance of SSA-LSTM, other different denoising methods are considered: Move Average filtering (MA), Median Filtering (MD) and Locally Weighted Scatterplot Smoothing (LOWESS). Meanwhile, the processed data obtained preprocessed by MA, MD and LOWESS are sent into Back Propagation network (BP), BP network optimized by Genetic Algorithm (GABP), Long Short-Term Memory network (LSTM) and Convolutional Neural Network (CNN) for comparison seperately. Mean Square Error (MSE) and coefficient of determination (R2) are used to evaluate the predicted performance. The experimental results show that the proposed SSA-LSTM has the highest accuracy for stock price prediction.