There is no hard and fast rule, but the choice depends on application. The application where smooth change over is required, MFs with smooth edges like Gaussian or Bell shapes are preferred.
There is no basci rule to identify the ANFIS MFs...the type and numbers of MFs are determined iteraively (usually..). Nevertheless, According to Russel and Campbell, Triangualr MFs are usually applied for practical applications. You might check som of my papers, where we have compared various MFs as well as their numbers...
Any way, you SHOULD keep the number of MFs at minimum to reduce the computational cost and efforts...
Thank you for your answer. Even I have noticed that as I increase the number of MF, it takes longer time to compute. Any particular paper you would like to recommend?
yes...the large the MFs numbers, the longer the ANFIS response...
By the way, you may check my following papers:
-Karimi et al., 2013 Neuro fuzzy and neural networks ...Computers and Geosciences,
-Kisi et al., 2013. Modeling rainfall-runoff process...Computers and Geosciences,
-Kisi and Shiri, 2012. River suspended sediment modeling...Computers and Geosciences
-Kisi et al., 2012. Forecasting daily lake levels...Computers and Geoscinces
In the mentioned papers, we have showed effect of MFs...Please see the list of publication in my page here. If you do not have access, please send a request for full text.