Comparative Study of CNN and LSTM on Short-Term Future Stock Price Prediction

Abstract

Stock market prediction is a practice of forecasting the company’s future stock values. A few years back, it was very challenging even for the expert analysts to project stock prices for various companies, but in the recent years, it has become easier to predict future stock values as people have started assigning more time to stock price prediction using various algorithms like SVM, LSTM, CNN, and RNN. So, in this context, an idea of comparative study of two frequently used ML algorithms like convolutional neural networks and long short-term memory is being suggested for stock price forecasting. To conduct this research, the network is trained by using the input data of previous years to predict the prospect value of the company’s stock and provide a graphical outcome that compares the actual prices and prices predicted using the model. The comparison of the two methods will be based on specific criteria that are required for obtaining a model. The major factors needed for implementation are open and closed pricing. Moreover, the RMSE, R2 score, and accuracy values for both models are compared for both open and closed prices in this study. The use of parameters will give us an overview on these two models which will define future global trade and economic growth of a country. Based on the results of the experiment, it has been observed that it is more reliable to use LSTM which gives an accuracy of 99.68% for open price and 99.89% for close price prediction than using CNN which gives 99.07% for open price and 97.89% for close price prediction.