2.3.1. Neural Network (NN) Algorithm
Neural networks predict a continuous or categorical goal based on one or
more predictors by identifying trends in the data which are unknown and
probably complex. A feed-forward, supervised learning network of up to
two hidden layers is the multilayer perceptron (MLP). The MLP network is
a feature of one or more predictors that minimize one or more target
prediction errors. A combination of categorical and continuous fields
can be predictors and targets (11). In the current study, the activation
function was a hyperbolic tangent, the error function was cross-entropy,
the number of hidden layers was two, and the number of component models
for boosting was ten. The boosting is an algorithm used to enhance model
stability/precision, can be used in all models, and can minimize
prediction variances and biases. Scaled Conjugate Gradient technique was
used for tuning model hyperparameters (12).