Stock Prediction Under GRU, LSTM and GNN

The stock market is not only a stage for investors to grow their wealth, but also an important indicator of a country's economic health and development. Deep learning techniques offer new possibilities for accurate stock market trend forecasting. The purpose of this paper is to explore the application of Recurrent Neural Networks (RNN), Long Short-Term Memory Networks (LSTM), and Gated Recurrent Units (GRUs) in stock market forecasting. By comparatively analyzing their performance, this paper aims to reveal which algorithm better captures complex fluctuations in the stock market. This paper shows that the gated recurrent unit (GRU) model exhibits superior performance compared to the LSTM and the RNN in the long-term stock price prediction task. In addition, it was found that in comparison experiments conducted with 100 and 500 training cycles, the LSTM model showed a significant performance degradation after 500 training cycles, revealing its potential shortcomings in handling long-term training tasks. These findings have important implications for understanding the adaptability and efficacy of various types of recurrent neural networks in complex prediction tasks.