Prediction of Stock Price by Neural Network Based on CNN, LSTM, ANN

Abstract

The stock market's complexity and volatility have long posed significant challenges for prediction models. This study delves into the comparative efficacy of three machine learning models, i.e., Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Artificial Neural Networks (ANN), in forecasting stock prices. Empirical evidence from a range of research papers is synthesized in the study, which emphasizes the advantages and limitations of each model when applied to the analysis of financial time series. CNNs, while adept at feature extraction, may not fully leverage the temporal dynamics of stock data. LSTMs, with their ability to capture long-term dependencies, have shown a propensity for enhanced prediction accuracy, yet they are computationally intensive. ANNs, despite their simplicity, offer a robust baseline for comparison but often fall short in capturing intricate market patterns. Based on the analysis, LSTMs generally outperform the other models, suggesting their suitability for complex financial forecasting tasks. The decision to select a model should be dictated by the demands of the forecasting endeavor, encompassing available computational assets and the preferred equilibrium between precision and intricacy. This investigation emphasizes the critical role that model choice plays in achieving dependable prognostications of stock prices and lays down a groundwork for future studies aiming to merge these models with a broader spectrum of financial metrics.