SAS PROC NEURAL Programs

SAS PROC NEURAL Programs

SAS Programs from Neural Network Modeling using SAS Enterprise Miner

The following is SAS programming code that is listed in Neural Network Modeling using SAS Enterprise Miner by fitting various predictive models using the SAS PROC NEURAL procedure.

MLP Neural Network Modeling

Finding the Best Weight Estimates in the Neural Network Model: SAS data mining programming code using the PROC NLIN non-linear SAS procedure by iteratively finding the best initial weight estimates in the MLP neural network model with one hidden layer by fitting an interval-valued target variable.

MLP Neural Network Model By Fitting an Interval-Valued Target Variable: SAS data mining programming code using the PROC NEURAL data mining procedure to fit an interval-valued target variable by incorporating the simplest MLP neural network model with only two hidden units.

MLP Neural Network Model With Two Hidden Layers By Fitting an Interval-Valued Target Variable: SAS data mining programming code using the PROC NEURAL data mining procedure fitting a MLP neural network model with two hidden layers to predict an interval-valued target variable.

MLP Neural Network Model Using an Exponential Activation Function: SAS data mining programming code using the PROC NEURAL data mining procedure by fitting the same MLP neural network model using an exponential activation function.

MLP Neural Network Model By Fitting a Ordinal-Valued Target Variable: SAS data mining programming code using the PROC NEURAL data mining procedure by fitting the simplest MLP Neural Network Model with two hidden units to predict an ordinal-valued target variable.

MLP Neural Network Model By Fitting a Nominal-Valued Target Variable: SAS data mining programming code using the PROC NEURAL data mining procedure by fitting the simplest MLP Neural Network Model with two hidden units to predict an nominal-valued target variable.

Generalized Linear Iterative Modeling

GLIM Model By Fitting a Ordinal-Valued Target Variable: SAS data mining programming code using the PROC NEURAL procedure to fit a GLIM neural network classification model with no hidden units to predict an ordinal-valued target variable.

GLIM Model By Fitting a Nominal-Valued Target Variable: SAS data mining programming code using the PROC NEURAL data mining procedure to fit a GLIM neural network classification model with no hidden units to predict an nominal-valued target variable.

GLIM Model By Fitting a Proportional Target Variable: SAS data mining programming code using the PROC NEURAL data mining procedure to fit a GLIM neural network classification model in estimating a proportional-valued target variable with an output activation function using a logistic transformation and a cross-entropy error function that is similar to the log-likelihood or deviance function with no hidden units.

GLIM Model By Fitting Multiple Proportional Target Variables: SAS data mining programming code using the PROC NEURAL procedure to fit a GLIM neural network classification model in estimating two proportional-valued target variables by incorporating a softmax activation function that restricts the values of the target proportional responses within the desired interval of zero and one and using the multiple entropy error function that is also referred as the log-likelihood function.

GLIM Model By Fitting a Multiple Target Variables: SAS data mining programming code using the PROC NEURAL procedure to fit a multivariate regression model to predict numerous interval-valued target variables.

GLIM Model For Robust Regression: SAS data mining programming code using the PROC NEURAL procedure by fitting the GLIM Model to predict an interval-valued target variable using the HUBER error function for robust regression modeling. Robust regression modeling is used when fitting models with extreme values.

Link-Linear Neural Network Model: SAS data mining programming code using the PROC NEURAL procedure to fit a link-linear model to predict an interval-valued target variable with non-negative values. Log-log link modeling is appropriate when the fitted values are also non-negative. In other words, the ACT=EXP output activation function transforms the expected target event into non-negative estimates.

Miscellaneous SAS Programming Code Using the PROC NEURAL Procedure

Scoring the Neural Network Model: SAS programming code using the PROC NEURAL procedure to find the best initial weight estimates based on the smallest error from the validation data set. The weight estimates are refitted into the subsequent MLP neural network model.

Input Pruning: SAS programming code that performs a variable selection routine called sensitivity-based input pruning in neural network modeling. The importance of input pruning is that it is a process that can be used in selecting the best set of input variables to the neural network model. In the following example, the input variable with the smallest Bernoulli deviance statistic is removed from the model. The Bernoulli deviance modeling assessment statistic is used since the neural network model is fitting a categorical-valued target variable.

Input Pruning or Growing: SAS programming code that performs input pruning based on the Schwarz's Baysian Criterion statistic in which two separate input variables are removed from the MLP neural network model. Input pruning is performed by commenting out the PROC NEURAL procedure code for the hidden layers.

ROC Area Statistic: SAS programming code that performs a modeling selection statistic called the ROC area statistic. The SAS programming code will calculate the ROC area statistic from the penalty SAS data set to compare the accuracy between the MLP neural network modeling probability estimates and logistic regression modeling probability estimates.

SAS Programming Code to Compare Neural Network Estimates with Traditional Statistical Estimates

The following is the SAS programming code that is listed in the appendix of Neural Network Modeling using SAS Enterprise Miner which compares neural network estimates with traditional statistical modeling estimates.

Comparing Traditional Regression Estimates: SAS programming code that generates scatter plots to compare traditional regression estimates with neural network estimates.

Comparing Logistic Regression Estimates: SAS programming code that generates scatter plots to compare logistic regression estimates with neural network estimates.

Comparing Regression Modeling with Autocorrelation Errors: SAS programming code generates scatter plots to compare regression models with autoregressive error estimates with neural network estimates.

Comparing Time Series with One-Step-Ahead Estimates: SAS programming code generates scatter plots to compare autoregression time series estimates with neural network estimates.

Comparing Discriminant Analysis Estimates: SAS programming code that generates numerous classification boundaries to compare the classification performance between discriminant analysis and many different neural network designs.

Back to Previous Page