Nidhi Sharma, Manisha Jailia
This study presents a hybrid deep learning framework that integrates graph convolutional networks with a chaotic long short-term memory (LSTM) model to improve stock market prediction. This approach captures both spatial and temporal dependencies among stocks by constructing daily similarity graphs using trading indicators such as price, volume, and volatility. These graphs reflect the structural behavior of the market and are processed through the GCN layers to extract spatially aware features. The sequential embeddings are then passed to a Chaotic LSTM, enhanced by the Chen chaotic system, to better model the non-linear and volatile time-series patterns. The model was trained and evaluated on datasets from NASDAQ, the Chinese stock market, and the Nifty 50. The performance was further tested using various weight initialization methods. The proposed model achieved a Mean Absolute Error of 0.0213 and a Mean Squared Error of 0.0011, demonstrating its effectiveness in learning complex market dynamics and improving the accuracy of future stock price predictions.