A novel method of bayesian genetic optimization on automated hyperparameter tuning.

Area of Science:

• Computational Finance
• Machine Learning
• Financial Forecasting

Background:

• Manual hyperparameter tuning in financial models can reduce prediction accuracy.
• Deep Neural Networks (DNNs) show promise but require careful parameter optimization.
• Symbolic Genetic Programming (SGP) offers a framework for automated model development.

Purpose of the Study:

• To develop an automated hyperparameter tuning method for stock market prediction using a novel Bayesian-based Genetic Algorithm (BayGA).
• To integrate BayGA with a Deep Neural Network (DNN) framework for enhanced financial forecasting.
• To evaluate the predictive performance of the proposed model against major stock indices.

Main Methods:

• Integration of Symbolic Genetic Programming (SGP) with Bayesian techniques within a Deep Neural Network (DNN).
• Development and application of a Bayesian-based Genetic Algorithm (BayGA) for automated hyperparameter optimization.
• Comparative analysis of the proposed DNN-BayGA model against benchmark stock indices (HS300, CSI500, CSI1000).

Main Results:

• The DNN model combined with BayGA demonstrated superior performance compared to major stock indices.
• Annualized returns for the DNN-BayGA model exceeded HS300 by 10.06%, CSI500 by 8.62%, and CSI1000 by 16.42%.
• The model achieved significant Calmar Ratios: 3.83 for HS300, 2.71 for CSI500, and 6.20 for CSI1000.

Conclusions:

• The proposed Bayesian-based Genetic Algorithm (BayGA) effectively optimizes hyperparameters for stock market prediction.
• The integrated DNN-BayGA framework offers a robust and high-performing solution for financial forecasting.
• This approach significantly enhances predictive accuracy and financial returns in stock market analysis.