International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 09 | Sep 2025
p-ISSN: 2395-0072
www.irjet.net
Hybrid Deep Learning Approaches for Stock Price Prediction Using LSTM and ARIMA Models P Sanyasi Naidu1, Beena Renuka2 1Professor, Department of Computer Science and System Engineering, Andhra University College of
Engineering, Andhra Pradesh, India.
2Student, Department of Computer Science and System Engineering, Andhra University College of Engineering,
Andhra Pradesh, India. ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract – This study presents a hybrid forecasting
advantages, LSTM models are prone to overfitting when trained on limited datasets and require significant computational resources.
framework that integrates Autoregressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM) models to improve stock price prediction. ARIMA is effective in identifying short-term trends and linear components in time series data, while LSTM excels in capturing nonlinear dependencies and long-range patterns. By combining these two approaches, the proposed model leverages the strengths of both statistical and deep learning methods. In the two-stage design, ARIMA first models the linear structure of the stock data, and the residual errors are subsequently processed by the LSTM to account for nonlinear fluctuations. Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R2) are among the error metrics used to validate the efficacy of the hybrid model. Experimental results show that the ARIMA–LSTM model consistently outperforms standalone ARIMA and LSTM, achieving lower error values and a high R² score of 0.9546. These findings indicate that the hybrid approach provides more accurate forecasts and demonstrates strong potential for practical applications in financial forecasting and decision support.
To overcome the limitations of using ARIMA or LSTM alone, researchers have proposed hybrid forecasting frameworks that combine statistical and machine learning approaches. Previous studies, such as ARIMA–ANN and ARIMA–neural network models, have demonstrated that integrating linear and nonlinear modelling techniques can lead to improved accuracy. More recently, the ARIMA–LSTM hybrid approach has gained attention for its ability to capture both short-term linear structures and long-term nonlinear dependencies in financial data. Motivated by these insights, this study develops and evaluates a hybrid ARIMA–LSTM model for stock price prediction. This work makes three contributions: • Creating a two-stage hybrid model in which LSTM models nonlinear residuals and ARIMA captures linear components. Using error metrics like MSE, RMSE, MAE, MAPE, and R2 to assess the model. • Demonstrating that the hybrid approach achieves superior accuracy, explaining over 95% of the variance in stock price movements.
Key Words: ARIMA, Deep Learning, Financial Forecasting, Hybrid Model, LSTM, Stock Price Prediction.
1.INTRODUCTION
2. METHODOLOGY
Forecasting stock market prices remains a complex challenge due to the highly volatile, nonlinear, and dynamic nature of financial data. Traditional econometric models such as the Autoregressive Integrated Moving Average (ARIMA) have been widely applied because of their ability to capture autocorrelation and linear dependencies. However, these models are limited in representing nonlinear behaviours that are common in real-world financial time series. On the other hand, deep learning techniques, particularly Long Short-Term Memory (LSTM) networks, have shown considerable success in modeling sequential and nonlinear patterns. LSTM, a variant of recurrent neural networks (RNNs), incorporates memory cells that enable the network to retain information over long sequences, making it suitable for learning long-term dependencies. Despite its
© 2025, IRJET
|
Impact Factor value: 8.315
This paper’s research approach blends the statistical soundness of the Autoregressive Integrated Moving Average (ARIMA) model and the non-linear learning Long Short-Term Memory (LSTM) networks. Its workflow includes six stages: data collection and preprocessing, ARIMA modeling, residual extraction, LSTM modeling, hybrid integration, and performance evaluation. Figure 1 displays the overall system framework. 2.1. Data Collection and Preprocessing Stock market data covering multiple years was collected from publicly available financial databases. The dataset contained daily values such as open, high, low, close, and trading volume, ensuring representation of diverse market
|
ISO 9001:2008 Certified Journal
|
Page 7