On stock price prediction: A deep learning approach using bidirectional long-short term memory (BiLSTM)

Abstract

Forecasting stock prices is a challenging topic that has been the subject of many studies in the field of finance. Using machine learning techniques, such as deep learning, to model and predict future stock prices is one approach to this problem. Bidirectional Long Short-Term Memory (BiLSTM) is a kind of recurrent neural network (RNN) that is particularly suitable to this task since it can accept input sequences of varying lengths and "memory" previous inputs. In this study, we utilize BiLSTM to forecast the stock price of Apple Inc. based on a training set including five-year records of Apple stock prices. The objective of this experiment is to examine the advantages of using BiLSTM for stock price prediction. In specific, we compare the BiLSTM model with other machine learning and deep learning methods including SVR, KNN, Vanilla RNN, LSTM, and show that BiLSTM is the best model in terms of six different performance metrics. In this work, we started with a description of the data preparation methods and the BiLSTM model design. We then examined the use of different optimization techniques like Adam, RMSprop, and SGD, as well as activation functions (Sigmoid, Tanh, and ReLU) in order to enhance the performance of the BiLSTM model. We later showed that Adam outperforms other optimization algorithms and that ReLU has the best performance among the three activation functions. The obtained results demonstrate that the BiLSTM model works best with the Adam optimization algorithm as well as ReLU activation. The BiLSTM model was also able to capture the data’s nonlinear and sequential patterns, resulting in more accurate predictions. Specifically, BiLSTM surpasses all conventional machine learning and deep learning methods taken into consideration.