Neural Network Model for Early Detection of Currency Crisis in Singapore: Optimization Comparison

This study aims to develop an early detection model for the currency crisis in Singapore using a neural network (NN) algorithm based on macroeconomic indicators. By utilizing the exchange market pressure (EMP) threshold approach, this study identifies currency crisis signals that can function as early warnings. In this study, a comparison of four NN optimization methods was conducted, namely stochastic gradient descent (SGD), Adam, Nadam, and AdaBound. The data used include 11 macroeconomic indicators from January 1990 to June 2021. The study results show that the NN model with Nadam optimization provides the best performance, with higher accuracy, sensitivity, and specificity compared to other optimization methods. This model successfully predicted crisis signals with an accuracy of 95.89%, a sensitivity of 98.36%, and a specificity of 83.33%. These findings can be used as a basis for decision-making in anticipating the currency crisis in Singapore.