Deep Learning in Time Series Analysis

5 Multi-Layer Perceptron (MLP) Neural Networks for Time Series Classification

In the functional point of view, a multi-layer perceptron neural network inherently performs a nonlinear regression with the difference from classical methods where the parameters associated with the method are found in an iterative way instead of analytic ways. Any iterative method needs terminating criteria, to terminate the recurssion, when the criteria are met, otherwise the method will make an over-fitting on the decision boundaries. The number of the epochs for training in conjunction with the classification error, are the two criteria employed in most of the applications. The architecture of a MLP contains an input node, constituted of the feature vector, followed ...