Adaptive distributed grid-partition in generating fuzzy rules

One common approach in constructing fuzzy classification systems based on if-then rules is using grid-partition. However, how to choose a precise size of the grid structure is still a problem. If the size of the grid is too coarse, the accuracy will be low, whereas if the size of the grid is too fine, rules could not be generated since there will be no object related to its membership function. To overcome the fine grid structure, a distributed-fuzzy-grid technique had been suggested. Yet this technique still has a problem in term of the number of generated-rules which increase significantly on every iteration caused by simultaneous overlapping grid-structure. In this research, an adaptive distributed grid-partition is proposed to generate fuzzy rule(s). Using this method, the number of generated rules can be adapted according to the needs so that the accuracy of the system as well as the complexity of inference time can be managed appropriately. The experimental results show that the average accuracy of rules generated using the adaptive distributed grid-partition is 96.59% compared to 96.36% resulted by the distributed grid-partition. In addition, an average accuracy of 98.83% was resulted compared to 97.166 obtained by another method.