Dear professor Aliyeh Kazemi

My mean is functional linear regression ,not fuzzy linear regression.

anyway thank you for your response.

It depends on what you are modeling.  For example, is this a physics experiment where you have reason to believe that the outcome should follow a smooth curve?  Then you will want to fit whichever trigonometric, polynomial, etc. functions.

I work at a hospital, where patient volumes follow seasonal patterns, and the changes are often abrupt.  So I have indicator (dummy) variables for month, day of week, etc.  Also, if a hospital added more beds or shut down a service on a certain day, there can be a sudden change in volume starting on that day.  I've found step functions like these usually fit the data better than linear functions.