A Proposed Hybrid GA-TDDPL-CNN-LSTM Architecture for Stock Trend Prediction

This study constructs a hybrid deep learning model to predict the price trend movement of Standard & Poor’s 500 index. Predicting stock market price trends is challenging because stock market data are non-linear and complex. Additionally, various factors, such as investor sentiment and news events, exert influence on stock price trends, leading to fluctuations in price trends. Researchers have implemented a variety of machine learning methods to predict stock price movements. The present study develops a hybrid deep learning network model consisting of a feature learning model, which is a long short-term memory model, and a feature selection model. Different types of data, including stock price, smoothing indicators, trend indicators, and oscillator indicators, are used as inputs to improve the model’s performance. Furthermore, to optimize the hyperparameter of each feature extraction model and feature selection model, a Genetic algorithm is utilized. An expert rule trend deterministic layer is also implemented to pre-process the data to further improve the model’s performance. The results indicate that the proposed model has superior testing performance compared to restricted Boltzmann machine, convolutional neural network, and autoencoder models.