A model auto-selection financial data simulation software using neuron-adaptive feedforward neural networks
Shuxiang Xu, Ming Zhang · 2003
In this paper, a model auto-selection (MAS) neural network financial data simulation software, MASFinance, has been developed for use on UNIX. The core of the software is a neuron-adaptive feedforward neural network (NANN) and a learning algorithm that combines steepest descent rule with a pruning method, and the software serves as a tool for selecting a near optimal neural network simulation model for economic data. Our test outcomes show that, for a given set of real life financial data, MASFinance can automatically choose a near optimal simulation model (a NANN with a near optimal neuron activation function and near optimal numbers of hidden layers and units) and then simulates the data with a root-mean-squared error of less than 5%.