An Intelligent Hybrid Gcn-lstm Model For Energy Stock Price Forecasting With Temporal Dynamics And Inter-stock Correlation

• Author(s):

  Mrs. Banuppriya P | Yokesh Kumar E | Ranjith Kumar S | Poovarasan M | Arun Kumar R

• Keywords:

  Stock Price Prediction, Graph Convolutional Network (GCN), Long Short-Term Memory (LSTM), Relative Strength Index (RSI), Dynamic Time Warping (DTW), Energy Sector, Deep Learning, Financial Time Series.

• Abstract:

  Energy Sector Stock Price Prediction Is A Critical Yet Challenging Task In Financial Forecasting, Characterized By High Volatility, Non-linearity, And Complex Interdependencies Driven By Geopolitical Events, Regulatory Shifts, And Commodity Price Fluctuations. Traditional Statistical Models And Standalone Deep Learning Approaches Fail To Simultaneously Capture Both The Temporal Dynamics Within Individual Stock Sequences And The Spatial Dependencies That Exist Between Correlated Companies. This Paper Proposes An Intelligent Hybrid GCN-LSTM Model Augmented With The Relative Strength Index (RSI) As A Domain-specific Technical Momentum Indicator. The Proposed Architecture Integrates A Graph Convolutional Network (GCN), Which Employs Dynamic Time Warping (DTW)-based Correlation Analysis To Construct An Inter-stock Graph Capturing Spatial Dependencies, With A Long Short-Term Memory (LSTM) Network Enhanced By An Attention Mechanism For Modelling Temporal Price Patterns. RSI Is Incorporated As An Additional Input Feature, Providing Overbought And Oversold Market Signals That Enrich The Model's Understanding Of Momentum-driven Price Reversals. Comprehensive Experiments Conducted On 30 Top Global Energy Sector Stocks Sourced From Yahoo Finance, Covering The Period From March 1, 2011 To September 26, 2024, Demonstrate That The Proposed GCN-LSTM+RSI Model Achieves A Mean Squared Error (MSE) Of 0.061, Root Mean Squared Error (RMSE) Of 0.247, Mean Absolute Error (MAE) Of 0.198, And A Coefficient Of Determination (R²) Of 0.872. These Results Represent A 21.8% Improvement In MSE Over The Base GCN-LSTM Model And Significantly Outperform Standalone LSTM, GRU, MLP, And Linear Regression Baselines.

---

Other Details

• Paper id:

  IJSARTV12I3104734

• Published in:

  Volume: 12 Issue: 3 March 2026

• Publication Date:

  2026-03-18

Download Article