Analysis of Forecasting Stock Prices Using CNN Model

Abstract

Creating a trading strategy and selecting the ideal time to purchase or sell stocks depends in large part on stock price expectations. This paper provides a CNN-based stock price time series forecasting method, which proves the optimality of the model by comparing the accuracy of different models, which provides a possible direction for the exploration of stock price forecasting. This paper first introduces the working principle of CNN, LSTM, and Conv1D, and then experiments are carried out by establishing a model, and finally the relevant conclusions are obtained. The experimental results show that the Trainscore RMSE, Train MAE, Testscore RMSE, Test MAE, and MAE of CNN has a smaller size. Thus, in comparison to the LSTM and Conv1D-LSTM, CNN is the model with the best efficiency and greatest accuracy in forecasting, which is more suitable for investors to predict future stock prices than LSTM and Conv1D-LSTM.