As evident, the performance of simple grid and mean and standard deviation is comparable. But the performance of histogram and modified grid approaches is not good enough with the other approaches. This is because in the histogram approach a single fuzzy rule is not enough for each class and the classification of some patterns was rejected and in the case of the modified grid approach the number of fuzzy if-then rules is increased exponentially with the dimensionality of pattern space. Simple grid approach gave the overall best results with a classification accuracy of 99.73%. Rule generation using mean and standard deviation is easy to implement as it depends only on the mean and standard deviation of the attribute values.
The modified grid approach did not produce the desired accuracy. Moreover, in the grid-based approach, the number of fuzzy if-then rules exponentially increased with the dimensionality of the pattern space. Thus, a large number of fuzzy if-then rules are usually generated for real-world pattern classification problems. This leads to several drawbacks: over-fitting training patterns, large memory storage requirement, and slow inference speed. On the contrary, the numbers of fuzzy if-then rules in the first two approaches are the same as the number of classes.