By Kardi Teknomo, PhD .

NN

Share this: Google+
< Previous | Next | Contents >

What is Neural Network?

Artificial Neural Network is a family of models that intended to mimic the neural cells in the brain. Artificial Neural Network is also part of biologically inspired computing. It is not the intention to model the exact mechanism of how biological the neural cells in the brain work nor to create biological clone. Instead, the biology is only the source of inspiration. The unit processing element is a simple model called a neuron. Each neuron is basically a function that can receive many inputs and produces only one output. Combining many neurons into a network of neurons is what we called as Artificial Neural Network or simply Neural Network for short.

Example

Time series data are abundant especially in Finance. Stock price, commodity price, foreign exchange rate, inflation rate, country growth rate, power consumption load and so on are a few examples of time series data. Suppose you have a time series data that represent daily foreign exchange of Japanese Yen to USD. The actual date is converted into integer number to represent time, which begin at certain date (that you specify) and end when your data end. The weekend is skipped thus, the time does not really represent 24 hour day but the day when the data exists.

Time

1

2

3

4

100

101

Value

100.96

100.94

100.93

100.93

 

96.83

97.03

 

The plot of the few time series data is shown below.

TimeSeries

How to use feedforward neural network to predict time series data? In this tutorial, you will learn how to train the neural network to solve this type of problem. You will learn complete steps from how to preprocessing the data, training, evaluation until prediction in Chapter 17. Forecasting from Time Series Data.

Example

Suppose we have the following 10 rows of training data. The training data is supposed to be a part of a transportation study regarding mode choice to select Bus, Car or Train among commuters along a major route in a city, gathered through a questionnaire study. For simplicity and clarity, we selected only 4 attributes. Attribute gender is binary type, car ownership is quantitative integer. Travel cost/km is quantitative of ratio type but in here we have convert them into ordinal type and income level is also an ordinal type. Our simple goal is to train a neural network to predict the transport mode of a person, given the four attributes of gender, car ownership, and travel cost and income level.

Attributes

Classes

Gender

Car ownership

Travel Cost

Income Level

Transportation mode

Male

0

Cheap

Low

Bus

Male

1

Cheap

Medium

Bus

Female

1

Cheap

Medium

Train

Female

0

Cheap

Low

Bus

Male

1

Cheap

Medium

Bus

Male

0

Standard

Medium

Train

Female

1

Standard

Medium

Train

Female

1

Expensive

High

Car

Male

2

Expensive

Medium

Car

Female

2

Expensive

High

Car

 

After the training of neural network using the above data, we can use the neural network for prediction. We can input any valid values that are not in the training data and the neural network can generalize the results of classification as shown in the picture below. You will learn how to train the neural network to solve this problem, complete from data preprocessing, training, evaluation until prediction in the application of neural network in Chapter 15. Classification Application: Transportation Mode Prediction.

Neural Network

Example

Problem of pattern recognition is something neural network can do best. Suppose we want to recognize optical characters such as letters and numbers. The sample of images of numbers that we would like the neural network to recognize (that we created these images using Excel) are shown below.

Image0

Image1

Image2

Image3

Image4

Image5

Image6

Image7

Image8

Image9

In this tutorial, we will show you the complete solution of pattern recognition of the image numbers above using neural network. You will learn how to create the images, train the neural network to get perfect solutions of synapses weights in chapter 16. Image Processing Application: Optical Number Recognition.

Example

From city demographic data, we have data of population (in hundred thousand) over 20 years (based on every 5-year census) as shown table. Can you use neural network to predict the population in 2020?

Year

1995

2000

2005

2010

2015

2020

Population (x10^5)

2.1

2.9

3.2

4.1

4.9

?

The complete numerical solution and the answer of this type of neural network using Microsoft Excel can be seen in Example 49)

Example

A car salesman wanted to use data and neural network rather than based on intuition. He started with collecting the data of one particular brand XYZ about the price and the age of used car (in number of months after government registration) and how long the car has been in used (in km). Can you help him to estimate the price of a new car of XYZ brand? A second hand car type XYZ has age of 21 months and have been driven for 14,991 km. How much is the estimate price?

Neural Network

The complete numerical solution and the answer of this problem using neural network is demonstrated in Example 50)

In this tutorial, you will learn a lot of numerical examples and spreadsheet solutions. The main focus will be feedforward network architecture. Especially, our discussion of neural network in this tutorial will include McCulloch and Pit, Perceptron, ADALINE, Multi-Layer Perceptron, MADALINE and Plug & Play Neural Network. In fact, the feedforward network models are the most popular ones that people often associate artificial neural network with these types of network architectures. The popularity of feed-forward network leads many people to (mistakenly) believe that feed-forward network is the only model of neural network. Of course, the actual neural network is a family of network models, not just feed-forward of network architecture.

For the learning methods, we unify the neural network learning from perceptron learning and Back Propagation training method. Other neural network training methods using Generalized Reduced Gradient (GRG) Non Linear optimization and evolutionary training methods in Excel Solver are also discussed briefly.

< Previous | Next | Contents >

Give your feedback and rate this tutorial

Read it off line on any device. Click here to purchase the complete E-book of this tutorial

See Also :
K means clustering , Similarity Measurement , Reinforcement Learning (Q-Learning) , Discriminant Analysis , Kernel Regression , Clustering , Decision Tree

This tutorial is copyrighted .

Preferable reference for this tutorial is

Teknomo, Kardi (2017). Neural NetworkTutorial. http:\\people.revoledu.com\kardi\tutorial\NeuralNetwork\