Stock Market Prediction Using Encoder Decoder ConvLSTM

Abstract

Stock market prediction is a hot topic these days and predicting the price of a stock is both difficult and important due to the numerous variables at play. There were numerous Machine Learning models used for Stock Market Prediction, but Hybrid Models were successful in making accurate predictions. The goal of this study is to create a hybrid Deep Learning model (Encoder-Decoder ConvLSTM) to predict stock market prices. We used historical stock price data from the Standard and Poor (S&P 500) from Yahoo's finance website. This dataset consisted of daily values with different features i.e., open price, high price, closing price, adj close, and volume. We also used a six-month dataset from the State Bank of Pakistan (SBP). The dataset included closing prices of different currency exchanges. In this research AUDUSD currency exchange has been used to predict the closing price of the next hour. Different prediction models have been tested on the S&P 500 dataset and their results have been compared with the proposed model, and the proposed model has been applied to the SBP dataset as well after it was discovered that it performed well on the publicly available dataset. To determine the effectiveness of the proposed model, we used the following performance metrics, root means square error (RMSE), mean absolute error (MAE), mean square error (MSE), and mean absolute percentage error (MAPE). Experiments indicate that the proposed model has the best performance metrics values when compared to other comparable studies, as the previous results for the S&P 500 dataset for RMSE, MAE, and MAPE using the traditional LSTM model was 16.471, 11.40, and 0.53. Similarly, in another study using the Deep LSTM model the experiments were performed on the same dataset, and the best results for performance metrics RMSE and MAPE were 8.416 and 0.143 respectively. In comparison with these results, the proposed model performed well and the results for RMSE, MAE, and MAPE which have been collected using the same dataset are 4.0471, 2.662, and 0.282. As a result, we can conclude that our model is suitable for the accurate prediction of the stock market in general.