For grid partitioning methods, because of equal segmentation of the input space, many unused and unnecessary rules are created. But for a minimal realization of (AN)FIS structure, the clustering-based approaches (i.e. genfis2 and genfis3) are better.
If you are using the anfisedit GUI tool, try the Grid Partitioning method for raw FIS generation. If you are training the ANFIS using anfis function and doing this programmatically, use the genfis2 and genfis3(FCM-based approah), to generate the initial FIS structure.
All of these are implemented and available to download, in the following FEX post:
