Neural Network Modeling

Duration: 2.0 days

This two-day course helps you understand and apply two popular artificial neural network algorithms, multi-layer perceptrons and radial basis functions. Both the theoretical and practical issues of fitting neural networks are covered. Specifically, this course teaches you how to choose an appropriate neural network architecture, how to determine the relevant training method, and how to construct custom neural networks using the NEURAL procedure.

Learn how to

• construct multilayer perceptron and radial basis function neural networks
• choose an appropriate network architecture and training method
• avoid overfitting neural networks
• perform autoregressive time series analysis using neural networks
• interpret neural network models.

Who should attend: Data analysts and modelers with a strong mathematical background

Prerequisites

Before attending this course, you should

• have an understanding of basic statistical concepts, which you can gain from the Statistics I: Introduction to ANOVA, Regression, and Logistic Regression course.
• have completed the SAS Programming 1: Essentials course or have equivalent knowledge.
• be familiar with SAS Enterprise Miner software. You can gain this knowledge from the Applied Analytics Using SAS Enterprise Miner 5.3 course.
• have completed a college-level calculus course.

Course Contents

Introduction to Neural Networks

• using the NLIN procedure for nonlinear regression
• using the REG procedure for polynomial regression
• using the GPLOT procedure for nonparametric regression

Network Architecture

• constructing multilayer perceptrons
• constructing normalized radial basis function networks

Learning

• statistical theory of error functions
• benefits and shortcomings of numerical optimization methods
• avoiding inferior local minima

The NEURAL Procedure

• input selection using weight interpretation
• input selection using sensitivity-based pruning

Hybrid Networks

• defining and illustrating the use of a counterpropagation network
• defining a generalized additive neural network (GANN)
• illustrating the use of the GANN paradigm and compare its performance against other methods
• defining and illustrating how a surrogate model can be used to understand a neural network's predictions

The AUTONEURAL Node (Self-Study)

• comparing autoneural architectures

Software Addressed

This course addresses the following software product(s): SAS Enterprise Miner.

This page was created using SAS software.