A fuzzy min-max neural network classifier with compensatory neuron architecture

This work proposes a supervised learning neural network classifier with compensatory neuron architecture. The proposed "fuzzy min-max neural network classifier with compensatory neurons" (FMCN) extends the principle of minimal disturbance. The new architecture consists of compensating neurons that are trained to handle the hyperbox overlap and containment. The FMCN is capable of learning data on-line, in a single pass through, with reduced classification and gradation error. One of the good features of FMCN is that its performance is almost independent of the expansion coefficient i.e. maximum hyperbox size. The paper demonstrates the performance of FMCN with several examples.