Stock Market Price Forecasting Using LSTM And GRU Networks

Abstract

Because of the inherent volatility and complexity of the financial markets, accurately predicting stock prices is a challenging task. By assessing the performance of many predictive models, including Random Forest, Stacked LSTM, Stacked GRU, ARIMA (Auto Regression Integrated Moving Average), network with long short-term memory (LSTM) and support vector machine (SVM), the study seeks to tackle this challenge. We thoroughly evaluate these model’s performance by looking at their capacity for generalization and prediction accuracy. The GRU model outperformed the LSTM model with a substantially lower validation loss of 0.0101 and a training loss of 0.0067, whereas the LSTM model obtained a validation loss 0.5048 with a training loss of 0.0130. The SVM with Support Vector regression (SVR) demonstrated a much larger Mean Squared Error (MSE) of 2878.20, whereas Random Forest model produced an MSE of 85.57. The GRU and LSTM models were further improved by the stacked architectures; the Stacked GRU achieved loss of 0.0361 and the Stacked LSTM achieved a validation loss of 0.9658. Despite being simpler, the ARIMA model yielded a Mean Squared Error of 11.8594. The results of this through investigation show that the GRU and Stacked GRU models perform the best