Neural Networks as Statistical Tools for Business Researchers

Abstract

Artificial neural networks are rapidly gaining popularity in the hard sciences and in social science. This article discusses neural networks as tools business researchers can use to analyze data. After providing a brief history of neural networks, the article describes limitations of multiple regression. Then, the characteristics and organization of neural networks are presented, and the article shows why they are an attractive alternative to regression. Shortcomings and applications of neural networks are reviewed, and neural network software is discussed.

Untitled

“…therefore, it is easy to use and understand as compared to the other statistical methods. it does not require general assumptions of logical analysis like the dissemination of data; impartiality of variables and dimensions of the sample except that error term e is normally distributed (Detienne et al, 2003). non-linear relationship.…”

“…Moreover, except the trial and error exercise, there is no standard way to select the number of units in a network for the rigorous training of the network (Detienne et al, 2003). Hidden layer processing is a black box by nature as the ann is best in solving linear or non-linear equations by training, but it cannot interpret the intention of a bond between input and output.…”

“…non-linear relationship. according to Detienne et al (2003), neural networks have the capability to deal with highly complex interactions more meritoriously than other statistical techniques. the ann architecture warrants superior performance in dealing with all degrees of nonlinearity.…”

“…software Packages the neural network by sas; sPss neural networks by iBM; neural network toolbox 3.0 (for MatlaB); Practical neural network recipes in C++ by Masters; and neuroshell 2 by the Ward systems group are among the most frequently used neural network software packages (Detienne et al, 2003). advantages of the ann Despite the ability of self-adjustment, the ann can manage incomplete, omitted or noisy input values; the ann does not entail any previous assumptions and mappings.…”

The Artificial Neural Network Method: A Practical Guide for Business Research

“…software Packages the neural network by sas; sPss neural networks by iBM; neural network toolbox 3.0 (for MatlaB); Practical neural network recipes in C++ by Masters; and neuroshell 2 by the Ward systems group are among the most frequently used neural network software packages (Detienne et al, 2003). …”