Training PSNN using Improved GA for Enhanced Closing Price Prediction of Natural Gas and Crude Oil

The intrinsic dynamic nonlinearity of the data makes financial time series forecasting problematic. To solve this problem, the current study suggests a PSNN-IGA hybrid framework, which integrates an Improved Genetic Algorithm (IGA) with a HONN known as PSNN. The goal of testing the model’s predictive capability is to anticipate one-day-ahead values of two major financial commodities, crude oil and natural gas. We use MAPE and MSE, two measures of accuracy, to evaluate the success of our forecasts. Under the same experimental settings, the PSNN-IGA is compared to multiple baseline models, such as PSNN-GA, PSNN-DE, PSNN-FFA, and PSNN-GD. In all commodities, PSNN-IGA consistently scored the lowest MAPE and MSE values, proving that it is effective at controlling uncertainty and capturing the complex patterns found in financial time series. Further validation of these findings was done using the Diebold-Mariano (DM) test, which statistically showed that PSNN-IGA’s predictive accuracy is significantly superior to the baseline models, in addition to standard error-based measures. This two-pronged validation, which employs performance measures in addition to statistical testing, strongly supports the model’s capacity to produce trustworthy and precise predictions in unpredictable energy markets.