Decomposition-Based Backpropagation Neural Network Ensemble Processing Model for Exchange Rate Volatility

Exchange rate volatility forecast is critical for both the government and the private sector when making monetary policy. Thus, the goal of this research is to find an appropriate prediction model for exchange rate volatility. This study employs a decomposition-based Backpropagation Neural Network (BPNN) ensemble model. This model collects data through decomposition and then models it with BPNN to provide a single network prediction. Each IMF uses the same parameters: a 4-4-1 input-hidden-output neuron network architecture, sigmoid and identity activation functions, a learning rate of 0.1, and the Flecher-Reeves network training function. To avoid overfitting caused by repeated parameter selection attempts, the same parameters are employed. Furthermore, predictors are chosen using the Principal Component Analysis (PCA) and Correlation Based Feature Selection (CFS) approaches, and Ensemble BPNN is performed based on the predictor findings. The currency rate data utilized is from the last two years (2021–2023), with a big data spread (469.46), indicating strong volatility. The EMD decomposition approach includes 6 IMF and EEMD 7 IMF. This study demonstrates that the decomposition based on BPNN Ensemble can decrease errors for both decomposition techniques with a performance percentage of 92.78<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> and 98.72% for MSE and MAE compared to a single NN, the selection of a single network that is more capable of reducing errors is CFS, and the method developed is capable of tracking the movement of exchange rate volatility.