High-frequency stock price prediction via deep learning

Abstract

We performed a comparative analysis of deep learning methods for high-frequency stock price prediction. Instead of directly analyzing one-dimensional stock price time series data, this study employs the Gramian Angular Summation Field method (Wang and Oates, 2015) to transform high-frequency stock prices into images, which are used to train ResNet models for prediction (hereafter referred to as the image-based prediction method). In addition, the same dataset (one-dimensional time series without image conversion) is used to train Artificial Neural Network(ANN), Long Short-Term Memory(LSTM), and one-dimensional convolutional neural network(1D-CNN) models, enabling a performance comparison with the results of the image-based prediction method.