Stock Price Forecasting Using LSTM And GRU

This project investigates stock price prediction using machine learning techniques to enhance forecasting accuracy. We assessed traditional algorithms such as Random Forest Regressor and Support Vector Regression (SVR), achieving Mean Squared Errors (MSE) of 424.01 and 232.65, respectively, with accuracies of 86% and 34.63%. To further improve performance, we implemented advanced models including Long Short-Term Memory (LSTM) networks, Stacked LSTM (S-LSTM), Gated Recurrent Units (GRU), and Stacked GRU (S-GRU), alongside the Autoregressive Integrated Moving Average (ARIMA) model. The LSTM and S-LSTM models attained MSE values of 71.79 and 97.05, with accuracies of 79.83% and 72%, respectively. The GRU and S-GRU models demonstrated MSE values of 88.69 and 25.49, with accuracies of 74.99% and 92.81%. The ARIMA model also performed effectively, with an MSE of 11.86 and an accuracy of 96.6%. The dataset used comprised 2417 entries with features including Adj Close, Open, High, Low, Close, and Volume. These results reflect notable advancements in prediction accuracy with the proposed models.

KEYWORDS: Stock price, Machine learning, LSTM, GRU, Mean square error, Forecasting Stock.