17

What is it?

In order to understand what neural network analysis is we need first to know what a neural network is. Essentially, a neural network is a computer program modelled on the human brain which can process a huge amount of information and identify patterns in a similar way that we do. Many neural networks also learn as they process like we do and as such they improve over time.

Neural network analysis is therefore the process of analysing the mathematical modelling that makes up a neural network. This analytics technique is particularly useful if you have a large amount of data. Because neural networks recognise patterns and learn to improve their recognition ability, their insights can help make predictions. These predications can then be tested and the results used to improve decision making and performance.

Neural networks are already widely used in industries such as banking, fraud prevention, medicine and manufacturing.

When do I use it?

This type of analysis can be applied to many different systems of data in a wide variety of fields. In business, neural network analysis can help you improve sales forecasting, customer research, and target marketing. Analysis of neural networks can also be helpful in streamlining manufacturing processes and assessing risk.

It can also be used to determine the effectiveness of a neural network’s ability to learn. Remember, a neural network is designed to mimic the learning and pattern recognition features of the human brain so the results of the analysis can be checked against the results interpreted by a human user to see how close they are. This can guide the ongoing development of the neural network making it more and more useful and more and more accurate.

Whether helping people to solve statistical problems or creating simulations of complex environments for testing, the analysis is an essential part of making a neural network increasingly useful. There is little doubt that as they advance and evolve, the uses for them will continue to grow.

What business questions is it helping me to answer?

Neural network analysis can help to forecast the future and process large quantities of data. It can help you to answer:

• What products are our customers likely to buy?
  
• How many products are we likely to sell, especially across a complete portfolio of products with cross-effects?
  
• What variables influence the buying decision of our customers?
  
• What is the optimal allocation of advertising expenses?
  
• Where do we have bottlenecks in our manufacturing process?
  

How do I use it?

First you need to decide what problem you are trying to solve using a neural network and then gather data for training purposes. Essentially you need to train the neural network to process data and decipher patterns so that the result is constantly improving as the network learns.

Ideally the training data set should include a number of cases, each containing values for a range of input and output variables. You will need to decide what variables to use, and which and how many cases to gather.

Initially your choice of variables will be guided by your own experience of the problem area you are seeking answers on. Let your intuition guide you regarding which input variables are likely to be influential. Include the ones you think have the most impact and test to see if you are correct. Re-test until you whittle down the variables.

If you want to know more about neural network analysis and how to use it you can explore the links at the end of this chapter. Alternatively there are software tools and providers that can help you. These software tools will break down the results of the neural network analysis thus allowing you to make very accurate predictions by presenting the data in an easily digestible and understandable format.

The software may also allow the user to test out the changes that the analysis recommends to make sure that the initial prediction holds true.

Practical example

Neural networks are already used to create models of the human body which allow healthcare professionals to test out the results of certain medical interventions before they are conducted in the real world. This is of course incredibly useful and potentially lifesaving. These simulations then provide additional information that can help doctors make the right decisions.

Google’s Science Fair grand prize was actually won by an American teenager who used neural networks to create an app that can accurately diagnose breast cancer in biopsy tissue 99 per cent of the time. With no medical training Brittaney Wenger created the app using a vast amount of different data points, and the neural network is able to learn and detect patterns that can’t be detected by the human eye. For years doctors have found it incredibly difficult to diagnose breast cancer based on a biopsy but Wenger’s program is set to change breast cancer diagnosis forever.¹

Tips and traps

Neural network analysis is a complex analytics methodology that normally requires the input from experts in neural network analysis as well as the use of specialist software.

Further reading and references

To find out more about neural network analysis see for example:

• http://mitpress.mit.edu/books/mathematical-methods-neural-network-analysis-and-design
  
• http://www.wisegeek.com/what-is-neural-network-analysis.htm
  
• http://metalab.uniten.edu.my/~chensd/courses/Neural%20Network%20in%20MATLAB.pdf
  
• http://www.statsoft.com/Textbook/Neural-Networks
  
• http://www.ijcsit.com/docs/Volume%205/vol5issue01/ijcsit20140501140.pdf
  

_________________

¹BBC Two Horizon, ‘Monitor Me’ narrated by Dr Kevin Fong (2013)