Financial time series prediction using artificial neural network based on Levenberg-Marquardt algorithm

This paper discusses the applications of Artificial Neural Networks (ANN) using Levenberg-Marquardt optimization algorithm for prediction of financial time series. ANN based on Levenberg-Marquardt training algorithm outperforms gradient decent, conjugate gradient and other algorithms that use the first order derivative of performance index to optimize ANN weights. Levenberg-Marquardt algorithm uses a second order derivative of performance index (curvature information on error surface) as a Guassi-Newton algorithm, but it approximate Hessian matrix by the Jacobian (gradient). Experimental results shows efficiency using ANN based on Levenberg-Marquardt algorithm