LSTM-GNN Synergy: A New Frontier in Stock Price Prediction

Aims/ Objectives: Stock price forecasting can be a complex task because of the aggressiveness and growth of financial data. This study presents a new approach combining LSTM and GNN that overcomes the problems of separate models in stock price forecasting. Rather than handling time or relations separately, as is typical in existing systems, the proposed model can do both simultaneously. In addition, the suggested model enhances forecasting accuracy and defines a framework that can be used for other financial prediction problems. In contrast with existing approaches, in this case, this model is based on time series and the interrelation between stocks, which significantly enhances prediction accuracy. Time-series data from real stock markets and historical and fundamental stock market data have been used to validate this model. The results prove that the hybrid model is superior to traditional machine learning models and standalone models. Lower values of RMSE, MAE, higher values of R² scores characterize this model. This model has potential application areas such as financial forecasting, algorithmic trading, and portfolio management.

Design: The design is experimental since the stock market data were used to assess the performance of the machine learning models.

Place and Duration of Study: The model was trained and evaluated using stock market data retrieved from Yahoo Finance, which was available from 2020 to the present.

Methodology: The study consisted of data preparation, where the stock prices were normalized. This was followed by the development of LSTM and GNN models. The two models were used jointly to develop a hybrid LSTM-GNN model. The models were fitted to a training data set and tested on the test dataset using performance measures such as root mean square error(RMSE), Mean absolute error(MAE), and R squared.

Results: The findings showed that GNN outperformed the LSTM model, which had lower RMSE. The hybrid LSTM-GNN model had the highest R2 score and the lowest prediction errors. The hybrid model managed to show an enhanced prediction of stock prices compared to the individual models; thus, the effectiveness of the hybrid model is confirmed.

Conclusion: The hybrid LSTM-GNN model for stock price prediction has provided an innovative approach by integrating sequential patterns (LSTM) and interdependencies among inputs (GNN). Future research should focus on improving the model and evaluating it against actual financial datasets to leverage its capabilities for financial prediction in a wider context.