Evolutionary hybrid neural networks for time series forecasting

Forecasting the trajectory of time series is notably challenging, primarily attributed to the intrinsic non-linearities and continually shifting dynamics present in financial markets. In this research initiative, we adopt a pioneering methodology by leveraging the capabilities of an evolutionary algorithm named Barnacle Mating Optimization (BMO) for the precision adjustment of weights and biases within an Artificial Neural Network (ANN). This intricate optimization process results in the development of a hybrid model, aptly named BMO+ANN. We put BMO+ANN to the test by utilizing it for forecasting the closing prices of two widely tracked currency exchange rates. In order to provide a comprehensive comparison, we also train the same ANN model using DE and PSO algorithms resulting two competitive models such as DE+ANN and PSO+ANN and we engage them for the same task. The performance evaluation is carried out using RMSE metric. Remarkably, the results conclusively demonstrate that BMO+ANN outshines DE+ANN and PSO+ANN in its ability to make more accurate forecasting, underscoring the effectiveness of the evolutionary BMO algorithm in tackling the complexities of exchange rate time series forecasting.