The most important data requirements are:

The basic procedure of model building is described in the AdvancedMiner in practice chapter (see the sections Classification model building and Approximation model building). The full specification of model settings contains also the elements of algorithm settings: General Algorithm Settings, Learning Algorithm, Neural Layers and Transformation Settings.

It is possible to terminate the model building process (by selecting the Terminate Process in the process context menu). If possible, the partially trained model can be accessed. A model obtained in this way can be used like any other normally created model.

The procedure of model testing is described in the sections Classification model testing and Approximation model testing, respectively.

Algorithm settings

General Algorithm Settings

Table 34.1. Feed Forward Neural Networks: General Algorithm Settings

Name	Description	Possible values	Default value
Automatic Data Transformations	if TRUE automatic transformations (e.g. replaceMissing, binarization) should be executed, false otherwise.	TRUE / FALSE	FALSE
bipolarCategoricalAttributesBinarization	if TRUE then categorical values are binarized to bipolar values, otherwise to unipolar values	TRUE, FALSE	TRUE
dataStandardization		TRUE, FALSE	TRUE
errorFunction	the function used to compute the neuron output error	sumSquared, cityBlock	sumSquared
maxNumberOfIterations	the maximum number of iterations of the learning process	non-negative integer numbers	1000
minErrorTolerance	the stop criterion for the learning algorithm: the maximum acceptable error value for the training data	real non-negative numbers	0.001
noise	the variance of noise added to the data; when the value of this variance is equal to or less than zero then noise is not added	real numbers from the interval [0.0,1.0]	0.0
outputLayerActivationFunction	the activation function for the output layer	logistic, hyperbolicTangent	hyperbolicTangent
randomization	decides whether to randomize the order of the training data in each iteration	TRUE, FALSE	TRUE
seed	the seed for random generator; when the value of this seed is equal to zero random generator is initialized with a random value	real numbers	0
validationErrorTolerance	the tolerance for the error calculated on the validation data; the learning process will be stopped when the value of current error calculated on validation data is greater than the value of the minimum ever calculated validation error multiplied by this tolerance	positive real numbers greater or equal to 1.0	1.0
validationWindow	the Interval of fitting error calculation on the validation data. When the fitting error calculated on the validation data is growing, the learning process will be stopped. When the value of this variance is equal to zero or the validation data is not specified then this option is turned off	positive integer numbers	10

Figure 34.4. Feed Forward Neural Networks: Algorithm Settings

Learning Algorithm Settings

Table 34.2.  Feed Forward Neural Networks: Learning Algorithm Settings

Name	Description	Possible values	Default value
Batch Learning	enables/disables the batch learning process (faster but less accurate)	TRUE, FALSE (if available)	FALSE
Learning Rate	the value of the learning rate	real numbers from the interval [0.0,1.0]	0.1
Momentum	the memory of the previous change of neuron weights	real numbers from the interval [0.0,1.0]	0.0

Figure 34.5.  Feed Forward Neural Networks: Learning Algorithm Settings

Settings for Neural Layers

The user can decide about the structure of the hidden layers. A special editor is launched after double-clicking on neuralLayers in the algorithmSettings sub menu.

Figure 34.6. Feed Forward Neural Networks: Neural Layers Settings

The activation function and the number of neurons for each layer can be chosen. New layers can be added and existing layers deleted. It is also possible to delete all the hidden layers. However, the number of neurons should never be set to 0 or less, as this can cause errors.

In addition to the settings specific to the Neural Network algorithm, the user can use Transformation Settings - to control the way of data transformation; these settings are described in the Transformation chapter.

Model statistics

There are no statistics for the Neural Networks model available in AdvancedMiner. It is only possible to visualize the history of the learning error rate.

Figure 34.7. Feed Forward Neural Networks: history of the learning error rate

The results of the Feed Forward Neural Networks module are: