To make it simpler, just focus on the effects that different amounts of randii have on your Fuzzy system. If you set radii as 1 for all inputs, then genfis2, makes two Guassian membership functions for each input with large amount for sigma, so the range of their influence on data is large. On the other hand, when you use a smaller value for radii, then genfis2 makes more Guassian membership functions for each input data with smaller sigma, and that's reasonable because when you have Guassian membership functions with small sigma you need more of them to cover the entire data range. And also more membership functions need more Fuzzy rules. As a result with lower amount for radii you will have more membership functions and more rules for your data.
You also need to consider that genfis2 does not give you a good result for any data, as it is usually used as an initial system to train ANFIS. So, you probably need to use ANFIS (if possible) or insert concepts based on your own knowledge of the data into the system to get a better result from that.
