Skip to main content
main-content

Über dieses Buch

Scope of this Text This text is intended to provide the reader with an introduction to the analysis of numeri­ cal data using neural networks. Neural networks as data analytic tools allow data to be analyzed in order to discover and model the functional relationships among the recorded variables. Such data may be empirical. It may originate in an experiment in which the values of one or more dependent variables are recorded as one or more independent vari­ ables are manipulated. Alternatively, the data may be observational rather than empirical in nature, representing historical records of the behavior of some set of variables. An ex­ ample would be the values of a number of financial commodities, such as stocks or bonds. Finally, the data may originate in a computational model of some physical proc­ ess. Instead of recording variables of the physical process, the computer model could be run to generate an artificial analog of the physical data. Since data in virtually any native form can be expressed in numerical format, the scope of the analytical techniques and procedures that will be presented in this text is es­ sentially unlimited. Sources of data include research work in a range of disciplines as di­ verse as neuroscience, biomedicine, geophysics, psychology, sociology, archeology, eco­ nomics, and astrophysics. An often fruitful approach to data analysis involves the use of neural network func­ tions.

Inhaltsverzeichnis

Frontmatter

Introduction

Abstract
This text is extended to provide the reader with an introduction to the analysis of numerical data using neural networks. Neural networks as data analytic tools allow data to be analyzed in order to discover and model the functional relationships among the recorded variables.
Edward J. Rzempoluck

1. The Simulnet Desktop

Abstract
Simulnet is a program that is designed to carry out mathematical operations on numerical data. In contrast with traditional spreadsheet or statistical software packages, Simulnet provides analytical power that is not generally available with such software. These functions are based on a set of neural network and genetic algorithm-based algorithms. Simulnet operations can be grouped into four general categories: Data transformation, analysis, visualization, and modeling. This introduction includes exercises that allow the reader to explore a sample of the functions that are available in each of these four categories.
Edward J. Rzempoluck

2. Data Analysis

Abstract
This section will introduce a variety of approaches to the analysis of data. The primary focus will be on the application of neural network-based techniques to the tasks of prediction, classification, and function approximation. This section will therefore begin by discussing the following neural network functions that are available in Simulnet.
Edward J. Rzempoluck

3. Acquiring and Conditioning Network Data

Abstract
Training a neural network involves a number of considerations in getting from the process to be modeled to the actual set of network training exemplars. These considerations include the following (Stein, 1993):
  • Data Specification: Deciding on what variables should be included
  • Data Collection: Collecting samples from the included variables
  • Data Inspection: Inspecting the data for characteristic and anomalous features
  • Data Conditioning: Preprocessing the data to extract features, correct for anomalies, or to reduce the volume of data
Edward J. Rzempoluck

4. A Data Analysis Protocol

Abstract
This section presents a protocol, a systematic procedure, that can be used as a guide through the steps involved in preprocessing and analyzing experimental data. In contrast with most other sections of this book, the present section is designed be applied to the reader’s own data.
Edward J. Rzempoluck

Backmatter

Weitere Informationen