A Hybrid LSTM-GRU Model for Stock Price Prediction

The dynamic variation of the stock market plays a crucial role in assessing a country’s economic power and development. Modeling the chaotic fluctuations in stock prices aids investors and traders in uncertain situations by evaluating market trends for investment decisions. Previous methods utilized Recurrent Neural Network (RNN) models for prediction. However, none of the previous works attempted to leverage the full range of technical analysis features and use Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) modules to capture short and long dependencies. Initially, we use machine learning to extract appropriate features for making predictions. Subsequently, the impact of each feature is adjusted proportionally to the weight matrix extracted from the neural network. We combined LSTM and GRU to capture long and short information horizons in stock sequences to mitigate limitations of prior methods, such as the difficulty in hyperparameter tuning. By combining the capabilities of these two networks along together, we present a model for predicting market fluctuations. Our proposed model was evaluated in a specified test market, achieving a 3% improvement in Mean Squared Error (MSE) over the previous best approaches.