Tutorials on Neural Network Forecasting

Artificial Neural Networks have become objects of everyday use ... although few people are aware of it. Their superior performance in optical character recognition, speech recognition, signal filtering in computer modems etc. have established NN as an accepted model & method. However, neural networks have not yet been established as a valid and reliable method in the business forecasting domain, either on a strategic, tactical or operational level. Following we present selected applications of NN which have demonstrated their applicability in specific scenarios.

Considering forecasting objectives, we must differentiate between predictive classification tasks where the forecasted values are class memberships or the probabilities of belonging to certain class, i.e. binary predictors, and regression tasks, where the forecasted value represent a single number of metric scale, e.g. sales, temperature, rainfall etc., as in regression problems. Following we will refer to this as forecasting as opposed to classification.

In forecasting applications, many classification problems also encompass the "prediction" of a categorical or nominal scaled variable. In order to distinguish the distinct modelling approaches and pre-processing required, we consider forecasting applications where the predictor is of metric scale being regression or point prediction based (in the demand planning area simply denoted as sales or demand forecasting). Consequently, a rise / fall-predictions as in financial modelling of buy-hold-strategies would receive consideration as under classification tasks due to their nominal predictors.

On the following pages we wish to gather introduction material as well as extended tutorials on the use of Neural Networks for forecasting. Despite various pages introducing to the use of neural networks in general, they often lack the connection to a specific application. Additionally, most software simulators may be run either in fully-automatic modelling - effectively a complete black box outputting various numbers as a forecast - or in configuration mode, offering almost unlimited degrees of freedom in modelling NNs for the forecasting ask at hand.

In the meantime, you may access a tutorial on "Business Forecasting with Artificial Neural Networks", held for the 2004 Institute of Business Forecasting's tutorial for forecasting practitioners.

Tutorial 'Forecasting with Artificial Neural Networks' at the 2005 IEEE Summer School in Computational Intelligence EVIC'05, 14.-16.12.2005, Santiago, Chile

This tutorial gives an extended introduction into 'neural forecasting', providing demos, hands-on exercises, tips and tricks in modelling. In addition, it includes a brief introduction to time series forecasting and ARIMA modelling, indicating the similarities between Mulitlayer Perceptrons and nonlinear AR(p)-Processes. However, only the slides are online, which consist of only a fraction of the tutorial, excluding the hands-on experiments. We applied a NN software simulator specifically designed to teach & understand NN forecasting in time series or regression based forecasting.

	EVIC'05 Tutorial Slides: Forecasting with Artificial Neural Networks  (6MB)
	EVIC'05 Tutorial Notes (3 slides): Forecasting with Artificial Neural Networks  (3MB)

In addition, the attendees are invited to download a CD containing 22+ demo-programs of leading Neural Networks software companies, along with the example datasets to start experimenting and working with Neural Networks straight away. The CD further contains professional documentation and information on the application of neural networks. Due to the large number of programs and the size of over 250MB they cannot be made available here. Additional copies of the 2004 software CD may be ordered [here]

Regression Tutorial 'Business Forecasting with Artificial Neural Networks' at the IBF Conference, 23 & 24.08.2004, Boston

Artificial Neural Networks have received increasing interest in corporate business, promising superior performance, reduced forecasting errors and consequently enhanced decisions in strategic, tactical and operational planning. Already, some 5% of all companies in electricity and consumer products market are using neural networks to gain competitive advantages over their competitors.

This tutorial gives an introduction into 'neural forecasting', providing demos, hands-on exercises, tips and tricks in modelling. However, only the slides are online, which consist of only a fraction of the tutorial, excluding the hands-on experiments. We applied a NN software simulator specifically designed to teach & understand NN forecasting in time series or regression based forecasting.

	Presentation of 2004 IBF Tutorial 'Business Forecasting with Artificial Neural Networks'
	Data of 2004 IBF Tutorial simulations [zip-file]

In addition, the attendees receive a CD containing 22+ demo-programs of leading Neural Networks software companies, along with the example datasets to start experimenting and working with Neural Networks straight away. The CD further contains professional documentation and information on the application of neural networks. Due to the large number of programs and the size of over 250MB they cannot be made available here. Additional copies of the 2004 software CD may be ordered [here]

Introduction to Neural Networks in JAVA

Introduction to Neural Networks in Java introduces the Java programmer to the world of Neural Networks and Artificial Intelligence. Neural network architectures such as the feedforward backpropagation, Hopfield, and Kohonen networks are discussed. Additional AI topics, such as Genetic Algorithms and Simulated Annealing, are also introduced. The pages don't focus on NN for forecasting, but give a generally introduction & overview. Unfortunately they are heavily add-infused and therefore less readable ... but taken from a book and free to access! So give it a try: http://www.heatonresearch.com/articles/series/1/

Neural Network Ressources

Most questions regarding general NN modelling may be answered looking at the excellent NN FAQ of the comp.ai.neural-nets newsgroup hosted by SAS W.S. Sarle

Backpropagation Learning algorithm and its derivatives are explained in detail on Donald Tveters webpage Backpropagator's Review

©

Home | Neural Associations | Neural Applications | Neural Data Sources | Neural Community | Neural Contacts | Neural Publications | Neural News&Events | Neural Software | Neural Tutorials | Neural Links | Forecasting Principles

©  2002-2005 BI³S-lab - Hamburg, Germany - All rights reserved - Questions, Comments and Enquiries via eMail -  [Impressum & Disclaimer]
The Knowledge Portal on Forecasting with Neural Networks @ www.neural-forecasting.com -