Journal Article

A novel method of bayesian genetic optimization on automated hyperparameter tuning

Kamaruddin, Norshaliza, Li, Qi, Sui Ki Khoo, Ariel, Peng, Chen, Zhang, Jia

2025

Overview

This paper presents a novel approach that integrates Symbolic Genetic Programming (SGP) with Bayesian techniques within a Deep Neural Network (DNN) framework. The primary contribution of this research is the introduction of a Bayesian-based Genetic Algorithm (BayGA), designed to automate the tuning of hyperparameters for stock market prediction. Prior studies have shown that manual hyperparameter tuning can negatively impact prediction accuracy. The proposed BayGA method effectively optimizes critical hyperparameters’ tuning, leading to enhanced predictive performance. Experimental results show that the DNN model combined with BayGA outperforms major stock indices, achieving annualized returns exceeding those of the HS300, CSI500, and CSI1000 by 10.06%, 8.62%, and 16.42%, respectively, with Calmar Ratios of 3.83, 2.71, and 6.20. These findings underscore the effectiveness of the proposed BayGA technique in developing robust financial forecasting models.

• Publisher: Nature Publishing Group UK,Nature Publishing Group
• Subject: 639/166 / 639/705 / Accuracy / Algorithms / Bayesian analysis / Deep learning / Heuristic / Humanities and Social Sciences / Mathematical models / Methods / multidisciplinary / Neural networks / Optimization / Performance evaluation / Science / Science (multidisciplinary) / Securities markets / Stock prices