Dynamic modular fuzzy neural classifier with tree-based structure identification

Inspired by a modular way of reasoning, we present a subsethood-product fuzzy neural classifier with a novel, robust and dynamic architecture, involving a main module and a number of submodules. The well-known classification and regression trees (CART) algorithm is employed as a fast preprocess of structure identification, which divides the input space into high certainty and low certainty regions, each representing a primary fuzzy rule. These primary fuzzy rules use a minimum set of attributes and are mapped onto the main neuro-fuzzy module. However, the patterns belonging to a low certainty primary rule get further split into a subset of secondary rules that use an extended set of attributes. Each such rule subset is mapped onto an expert-submodule, which gets activated only when a pattern falls into the respective low certainty region. In other words, we create a rule form of “if—then–if” conditional statement, where the first “IF” concerns the main module and the primary rule, while the second “IF” concerns the respective submodule and the secondary rule set. This dynamic resource-allocating model is optimized through a supervised learning procedure. Experiments in benchmark classification tasks prove that this architecture not only does reduce complexity and computational cost, which is its primary goal, but also offers fast and accurate processing during real-time operation. Moreover, it holds certain properties that make it ideal for soft computing applications of high dimension, especially those that adopt user-profiles or require partial re-training.