Exploring the Impact of Neural Network Architectures on Prediction Accuracy in Complex Datasets

Abstract

In this work, to forecast the stock market prices a deep-learning neural network has been employed. The standard statistical models that have been primarily used up until now appear to be inaccurate and time-consuming, and most importantly, they have failed to accurately predict the complex behavior of the data. There are different neural networks present, each with its pros and cons and applications; however, to get the desired results recurrent neural networks (RNNs) are chosen to forecast the stock prices. RNN when used in its classical form also shows some limitations, it has a problem of vanishing gradient, which affects the handling of data. Researchers proposed a modified version of this architecture which is called Long Short-Term Memory (LSTM) architecture. The financial dataset of Apple Inc. used in this work is downloaded from Yahoo Finance. To gauge the working of the model in forecasting Apple Inc.’s stock prices several metrics are used. The metrics include Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R-squared (R2).