Gaussian random fuzzy and nature-inspired neural networks: a novel approach to Brent oil price prediction

Abstract Given the volatile nature of oil prices in the wake of COVID-19 and the Russia-Ukraine war, the need for advanced prediction models is evident. The Autoregressive Integrated Moving Average model estimated through the maximum likelihood method with Marquardt-BFGS optimisation (ARIMA-BFGS) was used to select the relevant predictors for three different models: the Extreme Learning Machine (ELM), the newly introduced Evidential Neural Network for Regression with Gaussian Random Fuzzy numbers (EVNN-FUZZY) and an Artificial Neural Network fine-tuned with Particle Swarm Optimisation (ANN-PSO). Formal unit root tests, Augmented Dickey Fuller (ADF) and Phillips-Perron (PP) are used to test the stationarity of the Brent oil price before estimating ARIMA-BFGS. Evaluation measures such as root-mean-squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and coefficient of determination ( $$R^2$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:msup> <mml:mi>R</mml:mi> <mml:mn>2</mml:mn> </mml:msup> </mml:math> ) are used to assess the performance of the models. The study utilises a combination of traditional methods and neural networks to improve the accuracy of the Brent oil price prediction. ANN-PSO improves the predictive precision of ARIMA-BFGS by 65.30% for the training dataset and 88.72% for the testing sample. The incorporation of COVID-19 and the Russia-Ukraine war has improved the performance of EVNN-FUZZY. Governments, investors and producers can all benefit from these outcomes while making financial decisions. The findings of this study can be used by oil-exporting economies to guide their budgets, while oil-importing countries can use them to manage inflation.