Artificial Neural Networks Using Microsoft Excel For Windows 95

Abstract

Artificial Neural Networks have been researched now for decades. The standard method of implementing Artificial Neural Networks is by using C++, Fortran, Pascal, or other high level computer language to develop a system able to take a set of inputs and generalize to produce a satisfactory output. The property of generalization allows Artificial Neural Networks to respond reasonably to novel inputs. With the high level of computer expertise needed to program such a network, the power of Artificial Neural Networks has been restricted to those with the ability to spend weeks in developing code for such an application. The purpose of this project is to research the feasibility of constructing a Neural Network in ExcelTM for Windows 95TM. This will bring the power of Neural Networks to the average computer user with a working knowledge of Excel for Windows 95.

1. Introduction This section explains the general organization of this paper and discusses a brief outline of the main objectives of this project.

Section 2 contains a general background of Artificial Neural Networks. Explanations will be given to explain how Artificial Neural Networks “learn” and generalize for a given set of data and how that helps in solving complicated systems measurement and control problems. Section 2 also discusses the significant understanding and knowledge of programming and Neural Network fundamentals needed to use the traditional C++ Neural Network.

Section 3 discusses the methods used to program the Neural Network using VBA (Visual Basic for Applications) in Excel. The programming knowledge required to use this interface is minimal and the only computer experience needed is a good working under-standing of Excel for Windows 95.

Section 4 provides conclusions and information that demonstrate that this Neural Network in Excel for Windows 95 performs as well, if not better, than its standard C++ counterpart.

2. Introduction to Artificial Neural Networks 2.1 Why Neural Networks? One may ask, “Why do we need Neural Networks? Why can’t we derive solutions analytically?” The explanation is simple. When using a large data set, the computation time is astronomical. To give an idea of how much time is involved, please consider the following analogy: Given a small data set with a two dimensional input, one has the equivalent of solving two equations with two unknowns. Any person with a good knowledge of algebra could solve such a problem in a few minutes. A computer could solve the same problem in a few milliseconds.

¦ This research was supported in part by National Science Foundation grant EEC-9531378