An Optimized Deep Learning Model for Stock Price Prediction Using Bi-Directional LSTM with Multi-Inputs and Multi-Steps

Abstract

Predicting stock prices accurately is an inherently challenging task due to the dynamic and fluctuating nature of various influencing factors. However, with the advent and implementation of deep learning, achieving precise stock predictions has become feasible. This study employs Bi-Directional Long Short-Term Memory (LSTM) models to forecast the closing stock price of Tesla for the following day. Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) are selected as indicators to show methods' performance. A new variable has been created through open and close stock price. The original method uses close stock prices as the only input for prediction. In contrast, the modified method uses both the open stock price and the created new variable for calculating close stock price. Both methods' parameters are firstly trained and adjusted on training and validation dataset for their best performance. Finally, both methods are applied to test dataset and the value of both indicators depict their prediction performance.