Evaluation of PSO-BiLSTM method for stock price forecasting using stock price time series data (Case study: Iran Stock Exchange and OTC stock)

In recent years, with the increase in the penetration rate of the capital market, more people have invested in the stock market. Predicting the stock prices accurately with the least error can reduce investment risk and increase investment return. Due to nonlinear fluctuations, stock prices prediction is often described as a subject of nonlinear time series that is influenced by many factors. In this study, the bidirectional long short-term memory (BiLSTM) method for predicting stock prices is evaluated. In this regard, several machine learning techniques are applied to predict stocks prices using time series, and finally two deep learning methods including a recurrent neural network algorithm (LSTM) and a bidirectional neural network algorithm (BiLSTM) are implemented and their results are compared. Time series data of price characteristics including open, closed, high and low prices for the selected value stocks listed in Tehran stock exchange and the OTC, are used as a case study to implement the mentioned methods. Considering the evaluation criteria of RMSE and R-Square, the results of this study indicated that the combined PSO-BiLSTM algorithm, predicts the stock prices more accurately and has a better performance than the BiLSTM, LSTM, SVR, CART and MLP algorithms.