Automatic Configuration of an Artificial Neural Network for Fire Risk Analysis

Artificial Neural Networks (ANN) have been used as a methodology for develop
fire risk models, achieving good results. This methodology tries to imitate the
human brain, due to your capacity of acquired knowledge on artificial neurons
connections. One of issues related with the use of ANN is its configuration
parameters: number of hidden layers, number of neurons in each layer, type
of activation function, among others. One standard approach is to do some
experimental tests by the expert until find to get a good performance. Here,
an automatic formulation for configuring the ANN is adotped, where the
problem is addressed as an optimization problem. The optimal solution
is computed by using the MPCA-NN (Multiple Particle Collision Algorithm
– Neural Networks), finding the best architecture for a supervised ANN. This
tool uses the backpropagation algorithm; and calculates the learning rate,
the momentum rate, the activation function, the number of hidden layers and
neurons