In this session there was one slide in particular that peaked my interested and got me thinking about use cases. Just how good is good enough?
The following slide in particular raised some questions for me. If the these various tools have the degree of variance that the stud presents (HES wrong by greater than 25%, 39% of the time, Rem/Rate wrong by greater than 50%, 25% of the time), what is the use case for these labels?
While many tools seem to be doing well on predicting how pools of homes function, individual homes are a different story with wide variance. If you are one of the homeowner that receives a label that says your house is 50% worse than it is (which will include potentially millions of American homeowners if we roll any of these systems out), you probably don't really care that the label is right on average.