Forecasting Financial Commodities Using an Evolutionary Optimized Higher-Order Artificial Neural Network

The dynamic nonlinearity approach and data series make financial time series prediction difficult. This research suggests the hybridization of an improved Firefly Algorithm (IFFA) and a higher-order artificial neural network termed as pi-sigma neural network (IFFA-PSNN) to forecast the future value of two financial time series (FTS). Two real-time financial commodities—crude oil and natural oil—are used to assess the suggested model. We evaluated the IFFA-PSNN using mean squared error (MSE) for one-day-ahead forecasting. The IFFA-PSNN is compared to several forecasting models, including the basic Firefly Algorithm (FFA)-based PSNN (FFA-PSNN), genetic algorithm (GA)-based PSNN (GA-PSNN), differential evolution (DE)-based PSNN (DE-PSNN), and gradient descent (GD)-based PSNN (GD-PSNN). The results show that the IFFA-PSNN model has the lowest MSE and very well integrates crude oil and natural gas data uncertainty.