Stock Price Prediction Using Graph Convolutional Recurrent Neural Networks: An Unified Approach for Classification and Regression

Traditional forecasting models such as ARIMA and ANNs fail to capture sequential dependencies and non-linear patterns of the market. Nevertheless, deep learning approaches offer LSTM and GRU methods to enhance sequential modeling but fail to establish complex inter-relationships within the financial domain. Therefore, in this work, we propose a framework titled Graph Convolutional Recurrent Neural Network (GCRNN) for merging spatial and temporal learning processes by introducing strength in financial forecasting. Graph convolutional layers capture interdependencies between financial indicators, while GRUs have shown their efficiency in modeling sequential patterns. The architecture integrates dropout layers to avoid overfitting and fully connected layers for enhanced contextual learning. Experimental results demonstrate a clear advantage over benchmark models-the GCRNN yields superior classification and regression results with 96.92% accuracy in classification, achieving a much lower predictor error in regression tasks over benchmark model approaches. From our results, we expect that GCRNN can extract strong and meaningful market signals and works as a scaling and adaptive solution for stock price prediction.