Recurrent neural networks for Apple stock price prediction

Abstract

In recent years, there has been a significant focus on exploring the application of neural network architectures for financial prediction. This present study investigates the utilization of a Long Short-Term Memory (LSTM) model trained on both quarterly fundamental data and daily historical stock price data of Apple (AAPL). The study evaluates the accuracy of different LSTM model variations trained on 29 different fundamental indicators using the Mean Squared Error (MSE), Root Mean Square Error (RMSE), MAE (Mean Absolute Error) and Mean Absolute Percentage Error (MAPE) in predicting stock future stock prices. The results show that by selectively choosing the fundamental indicators for training the LSTM model based on fundamental analysis, it can achieve a higher accuracy in comparison to a LSTM model trained exclusively on historical price data.