Genetic Algorithm based feed forward neural network for probability and predictive analysis of large data sets

 

Genetic Algorithm based feed forward neural network for probability and predictive analysis of large data sets.

I’ve been working on AI related systems for 10 years with expert systems and genetic algorithms.  I’ve recently gotten more interested in their application to neural networks and statistical analysis capabilities.

Basic background you need to follow this:

https://en.wikipedia.org/wiki/Neuroevolution

Basic Feed forward neural network:

https://en.wikipedia.org/wiki/Feedforward_neural_network

Basic Feed Forward Neural Network

Proposal:

We will develop a feed forward network that is self-tuning based off of an evaluation of it’s predictive results. i.e. fitness function will be used to evaluate connections/output on each iteration.

Applications:

The system could theoretically be utilized on any non-linear problem that requires an output at a given point of time regardless of accuracy.

https://en.wikipedia.org/wiki/Nonlinear_system

i.e. any type of complex analytical prediction, stock market, weather, migration data, resource utilization and conservation, conflict analysis and prevention.

Neural network probability estimation is widely researched and has a very good foundation.

https://www.google.com/webhp?sourceid=chrome-instant&rlz=1C1CHFX_enUS603US603&ion=1&espv=2&ie=UTF-8#q=neural%20network%20probability%20estimation

 

Output of the neural network could be utilized as either a tool for humans evaluate a situation/data i.e. “Projected Earnings” of a company, or they could be plugged strait into action oriented architecture.  i.e. Sell the Stock/Don’t sell the stock. 

Basic Example of an application.

 Weather data prediction.  This is most likely a system that has already been developed and probably works something like this:

Input Data Set is obtained via sensors in various locations:

Location

Barometric Pressure

Date/Time

Weather Condition

Wind speed

Wind direction

New York

5

1/1/2015 01:00

Sunny

35

East

Baltimore

10

1/1/2015 01:00

Cloudy

50

East

The time data would feed into the NN as it comes in and would be as frequent as possible.  The output of the neural network would be “Day Ahead” forecast per location.

The day after the “Day Ahead” forecast the network would be scored and the NN algorithm adjusted via the fitness function of each connection.  They would be reinforced or degraded as a result of its forecast accuracy.

 

 

Add new comment

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.
CAPTCHA
This question is for testing whether or not you are a human visitor and to prevent automated spam submissions.