Enhancing Stock Price Forecasting With Hybrid Ann-Ga Models: A Comprehensive Evaluation

Abstract

Accurate stock price prediction is vital for informed financial decision-making, yet it remains a challenging task
due to the inherent complexities and dynamics of financial markets. Traditional Artificial Neural Networks (ANNs)
have shown promise in forecasting stock prices by identifying patterns in vast datasets. However, their
performance is often limited by issues such as overfitting and local minima in optimization. This research
addresses these limitations by integrating Genetic Algorithms (GAs) with ANNs to improve predictive accuracy.
The ANN-GA hybrid model was developed and tested on stock market data.The model employed a learning rate
of 0.0015 and utilized a three-layer architecture with 64, 32, and 16 neurons, respectively. Genetic Algorithms
were used to optimize hyperparameters, resulting in significant improvements. The ANN-GA hybrid model
achieved a Mean Squared Error (MSE) of 0.0034, compared to 0.0058 for the traditional ANN model. The Root
Mean Squared Error (RMSE) improved from 0.0762 to 0.0583, and the model demonstrated a 24.6% increase in
accuracy. Additionally, the hybrid model reduced training time from 5 hours to 3 hours despite an increased timelag
of 5 days. These findings underscore the effectiveness of integrating GAs with ANNs, offering a more accurate
and efficient approach for stock price prediction. The research contributes to advancing predictive modelling
techniques in financial markets, providing valuable insights for traders and investors.