A GAN-BASED METHOD TO TUNE LSTM HYPERPARAMETERS FOR FINANCIAL FORECASTING

Optimizing hyperparameters is a critical challenge in enhancing the performance of Long Short-Term Memory networks for financial time series forecasting. Traditional optimization techniques such as grid search and random search are often computationally expensive and inefficient, while Bayesian optimization, despite its advantages, can struggle with exploration in complex search spaces. This paper introduces a novel Generative Adversarial Network-based approach to LSTM hyperparameter optimization, specifically applied to forecasting the next closing price of the S&P 500 index. The proposed method consists of a generator, which suggests potential hyperparameter configurations, and a discriminator, which evaluates their effectiveness based on forecasting accuracy. Through iterative adversarial training, the generator refines its suggestions, dynamically adapting to the optimization landscape and effectively balancing exploration and exploitation. The performance of the GAN-based optimization approach is evaluated using metrics such as Mean Squared Error, execution time, and resource utilization. Experimental results demonstrate that the proposed approach achieves competitive accuracy while improving efficiency and robustness in navigating the hyperparameter space. The findings of this study provide valuable insights into the application of adversarial learning for hyperparameter tuning, offering a promising alternative for enhancing LSTM-based financial forecasting models, particularly for the S&P 500 index.