Computational Intelligence: Synergies of Fuzzy Logic, Neural Networks and Evolutionary Computing

Appendix G

MATLAB® Programs for Neuro-Fuzzy Systems

The modelling approach used by ANFIS is similar to many system identification techniques and can be broken down into the following steps:

• Set of input/output data,
• Parameterized model structure relating to input/output MFs and rules.

In some cases, data is collected using noisy measurements, and the training data cannot be representative of all the features of the data that will be presented to the model. This is where model validation and testing come into play. The whole model-building process is divided into three steps:

• Model building,
• Model validation and
• Model testing.

Model validation is the process by which the input vectors from testing the I/O data set are presented to the trained ANFIS model to see how well the ANFIS model predicts the corresponding data set output values. To perform the above tasks, the whole data set is divided into three sets of data:

• Training data,
• Testing data and
• Checking data.

To create a training set from the available historical sequence first requires the choice of how many and which delayed outputs affect the next output. Each item in the training data set should have a value, because that is what the classifier uses to learn how to predict. In the checking data, each item may or may not have a correct value specified for the class value. To evaluate how accurate the classifier is, the true class values or checking data are needed. The classifier won't use them when making predictions, ...