5.11.4. Neural Network & MultiNomial Logistic Prediction (clip0243 action)

<< Click to Display Table of Contents >>

Navigation:  5. Detailed description of the Actions > 5.11. R Predictive >

5.11.4. Neural Network & MultiNomial Logistic Prediction (clip0243 action)

 

Icon: ANATEL~4_img15  

 
Function: R_NNET
 

Property window:

 

ANATEL~4_img14

 

Short description:

Compute a Neural Network Model or a Multi Nomial Logistic Model

 

Long Description:

Compute a Neural Network Model or a Multi Nomial Logistic Model

 
Neural Networks are popular in datamining and analytics, mainly thanks to some of its promoters (i.e. Google) and their increased popularity in image pattern recognition. There now exists a new type of Neural Network algorithms (that is named “deep neural network”) that seems to perform reasonably well on image classification and segmentation tasks.

 

The ANATEL~4_img15  action described in this section is not a “deep” neural network: it’s an “old-school” Neural Network algorithm and it’s included in Anatella mainly because of completeness (and for explanatory/teaching purposes). More precisely, “old-school” Neural Network algorithms are usually not very useful because they are notoriously difficult to adjust properly to get a correct classification accuracy (although it’s sometime possible to get good results, it’s quite difficult).

 

Parameters:
 

List of Predictors: Select independent variables

Target: Select the variable you want to predict

Model Output: Set the file name for the model results

Select Classification Model: either Multinomial Logit or Neural Network

Base: set the base category

Number of perceptrons for Neural Networks: Manually set the perceptrons. This implementation uses only one layer

Linear Model: Specify if the perceptrons should use linear models instead of logistics

Softmax: Switch to Log Likelihood criteria for more than two categories in the target