Several rule induction methods are available. Most of them use a FIS, the ``FPA'' algorithm, the "Wang & Mendel" algorithm and ``fuzzy decision trees''. In that case, existing rules, if there are any, are ignored, and MFs are not modified. The HFP method is a particular case. It is a partition refinement algorithm, described in the select a partition option (see section 3.1.2), which only uses data from a data file. If there is a current FIS, it is ignored by the method. The OLS method can use a FIS configuration, which means that the input fuzzy partitions are kept as such, it can also work from a data file only. In that case, one step is needed to generate the input fuzzy partitions.
Simplification can be applied to any FIS, independently of the induction method used to build it. It does not modify MFs, but only changes the rules.
Optimization can be applied to any FIS, independently of the induction method used to build it. It can modify all FIS elements including MFs.
We will specify whether there must be a current data file, and/or a current FIS, prior to the learning procedure. If these conditions are not fulfilled, the corresponding learning option will be grayed.
Note: most learning methods use a tolerance threshold EPSILON=. Before using these methods, it is better to normalize the data between 0 and 1, if their range is very high or very small.
Note: for all learning procedures and numerical outputs, the accuracy index is the RMSE performance index.