Higher Order Neural Networks with Bayesian Confidence Measure for the Prediction of the EUR/USD Exchange Rate

摘要: Multi-layer perceptrons (MLP) are the most common type of neural network in use, and their ability to perform complex non-linear mappings tolerance noise data is well documented. However, MLPs also suffer long training times often reach only local optima. Another higher-order (HONN) where joint activation terms used, relieving task learning relationships between inputs. The predictive performance tested with EUR/USD exchange rate evaluated using standard financial criteria including annualized return on investment, which shows a 8% increase compared MLP. output networks that give highest each category was subjected Bayesian based confidence measure. This improvement may be explained by explicit parsimonious representation high order networks, combines robustness against typical distributed models together accurately model interactions for long-term forecasting. effectiveness measure examining distribution network's output. We speculate could taken into account during training, thus enabling us produce properties take advantage