HYBRID NEURAL FUZZY SYSTEMS

Another use identified in fuzzy concepts is to combine them with neural networks. The standard neural fuzzy hybrid system is based on the inputs into a neural network being transformed into truth values at the input layer. This is a relatively unexplored area in effort prediction. The theory behind the system would indicate that neural nets would receive more crisp and meaningful inputs thus improving the overall output and quality of neural net predictions. An example of this hybrid system is presented in figure 8.

 

Figure 8 Diagram of Neural fuzzy system

In figure 8 there are two inputs ( data model size and process model size) which are assigned truth values at the first input layer. The output to the second layer represents the membership degrees in each valued logic (small, medium, large) which can lead to the rules layer where the rules represent the input weights. The output from this layer indicates whether the rule ‘fires or not’, determining the activation level at the output membership layer. Finally, the results are combined to give a single effort predication time.

As with ordinary neural nets the net learns on a training set and once the system has been trained it is possible to remove rules to check for acceptability. This is a very important feature as the network could learn undesired relationships from new data, ultimately forgetting old relationships from old data.

 

Assessment of hybrid neural nets.

 

 

This page was last updated 12/3/97 Dan. Snell, Bournemouth University, Copyright. 1997