Prediction of Index Trend based on LSTM Model for Extracting Image Similarity Feature

Abstract

The prediction of stock trend in the financial market has become an important application field in machine learning research. The fluctuation of stock price is affected by various factors and strong nonlinear and stochastic, which brings greater challenge to predict the future trend of stock. This paper demonstrates how to predict the index trend more accurately so as to bring more profits to investors. A large number of researchers in the financial field have tried to use machine learning algorithms, such as Support Vector Machines (SVM), Random Forests (RF), and Deep Neural Network (DNN) like Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) Network models to seek the dependence on the future trend of stock and historical stock data. This paper attempts to use Convolutional Autoencoder (CAE) to extracts the image implicit feature of the graphical representation to index data, and calculate the cosine similarity between other implicit features of its historical images. The image similarity feature is equal to the total sum of cosine similarity products by multiplying the corresponding trend direction value. Finally, we combine the image similarity feature and the filtered data features by Pearson product-moment correlation coefficient (PPMCC) as the input variables of the LSTM network model. After discussing the results of the comparative experiments, the prediction model with incorporating the image similarity feature has been improved on multiple classification evaluations.