CNN-GRU-XGBoost stock price prediction model under hyperparameter-based optimisation

This paper addresses the limitations of existing stock price prediction methods, which often lack explanatory power and struggle with complex hyperparameter combinations. We propose a hyperparameter tuning approach based on the CNN-GRU-XGBoost model. By employing Bayesian tuning for the CNN-GRU and random search for XGBoost, we identify the optimal hyperparameters. The CNN extracts local features from the data, while the GRU captures long-term dependencies. The combined features are then input into the XGBoost model for accurate stock predictions. Testing on five years of Hang Seng Index data demonstrates significant improvements in prediction accuracy, reduced noise, and enhanced model interpretability compared to traditional single models.