The fuzzy rules that were generated were ``flat'', in the sense they were all applied simultaneously. While this makes more sense with fuzzy rules than with standard bivalent rules, the natural extension is to take a cue from decision trees and attempt to construct fuzzy decision trees. These would be identical to standard decision trees except that the decisions at each branching point would be fuzzy, rather than bivalent. This makes calculating the best split somewhat harder (continuous functions for the information gain must be maximized), but would be more rewarding by allowing smaller trees with less leaves and internal nodes to encapsulate a richer amount of information.
Several papers have been written on this topic; [Boyen, 1995] describes the application of fuzzy decision trees to assessing the stability of power systems, while [Heinz, 1995] describes a hybrid method of adaptive fuzzy neural trees, combining fuzzy logic with both neural networks and decision trees.