Forecasting currency exchange rates with an Artificial Bee Colony-optimized neural network

摘要

This paper applies a recent variant of Artificial Bee Colony (ABC) to optimize the weights of a three-layer feedforward neural network for forecasting of currency exchange rates of USD/EUR and USD/Yen. The inputs to the network is built from historical prices and a set of well-known technical indicators, including Moving Average, Moving Average Convergence/Divergence and Relative Strength Index. The forecasting model becomes a complex minimization problem with fifty-decision variables, many of which are interdependent. The ABC variant in this work is ABCDE that is a hybrid algorithm of original ABC with two different mutation strategies of Differential Evolution (DE). The experimental results present a superior performance of ABCDE in terms of both training and testing errors against original ABC, Back Propagation and ODE [35], an efficient variant of DE.