BDMDM Term Paper Artificial Neural Network IIM Bangalore

Contents

Artificial Neural Network – An Introduction 3

Neural Network vs. Conventional Computing 3

Basic working method of Neural Network and learning process 4

Basic Neural Network Functions 6

Neural Network Models and Architectures 7

Learning Paradigms 7

Neural Network Topologies 8

Neural Network Models 9

Applications of Neural Networks 11

Neural Networks in Medicine 11

Marketing 12

Financial Domain 14

Technology Domain 15

Other Interesting Applications 15

Conclusions 17

Reference 20

Artificial Neural Network – An Introduction

“The human brain contains roughly 1011 or 100 billion neurons. That number approximates the number of stars in the Milky Way Galaxy, and the number of galaxies in the known universe. As many as 104 synaptic junctions may abut a single neuron. That gives roughly 1015 or 1 quadrillion synapses in the human brain. The brain represents an asynchronous, nonlinear, massively parallel, feedback dynamical system of cosmological proportions.” (Kosko (1992), pp 13)1

It is the human brain that acts as the model for Artificial Neural Network (ANN). ANNs are a class of powerful, general purpose data mining tools applied for the purposes of prediction, classification and clustering. They have been used across a broad range of industries, from predicting time series in financial world to diagnosing patients, from identifying customer segments to bankruptcy prediction, from recognizing hand written numbers to predicting failure rates of engines2.

ANN is an interconnected group of artificial neurons (programming constructs that mimic the properties of biological neurons) that uses a mathematical or computational model for information processing based on a connectionist approach to computation. In most cases, ANN is an adaptive system that changes its structure based on external or internal information that flows through the network. Essentially they are non-liner statistical data-modeling or decision making tools. They can be used to model complex relationships between inputs and outputs or to find patterns in data3.

In this report, we first explore some basic working method and structures of neural networks which will be useful in understanding their application. Next we discuss the various applications of neural networks in various industry or functional areas such as finance, marketing, medicine etc. In the last section, we analyze what are the advantage and disadvantage of neural networks in data mining field and when they are most suitable to be applied.

Neural Network vs. Conventional Computing

Neural networks take a different approach to problem solving than that of conventional computers. Conventional computers use an algorithmic approach i.e. the computer follows a set of instructions in order to solve a problem. Thus it restricts the problem solving capability of conventional computers to problems that we already understand and know how to solve6.

Neural networks process information in a similar way the human brain does. The network is composed of a large number of highly interconnected processing elements (neurons) working in parallel to solve a specific problem. Neural networks learn by example. Thus examples must be selected in such a way that the entire range of input or output specific to the business problem at hand is addressed. The disadvantage is neural network works like a black-box – we don’t know how the network arrives at a decision.

Neural networks and conventional algorithmic computers are not in competition but complement each other. There are tasks are more suited to an algorithmic approach like arithmetic operations and tasks that are more suited to neural networks. Even more, a large number of tasks, require systems that use a combination of the two approaches (normally a conventional computer is used to supervise the neural network) in order to perform at maximum efficiency.

Basic working method of Neural Network and learning process

Neural network consists of basic units that mimic, in a simplified way, the working of biological neurons in animal brains. In our brain, there are a large number of neurons or nerve cells that are connected to one another – the joining point is called synapse. When a signal passes through the neuron as well as the synapse, some amount of processing is done and essentially one or more input signals are combined to produce one output signal. As we do a job again and again and learn from our mistakes, essentially the internal nervous system gets trained at how to behave when a set of inputs is received.

Artificial neuron is modeled on the biological neuron. It is basically a programming unit with many inputs and one output. Each input has its own weight. In addition there is a weight called bias for the neuron. The neuron combines these weighted inputs and then transforms as per the transfer function to produce the output. The transfer function can have various forms – such as logistics, hyperbolic tangent etc. The output is thus a non-linear combination of its inputs.

The following are some illustrative diagrams for neural networks:

Inputs Outputs

Fig 1: Simple neural network. The result of training this network is equivalent to logistic regression

Input Hidden Layer Output

Fig 2: Neural network with hidden layer which enables the network to recognize complex patterns. [This is also called Multi Layer Perceptron (MLP)]