Although traditional system identification methods based on model class/parameter mode include almost all
the results of system identification and the theory is also mature, it is not appropriate to the relatively less data which is
uneven distributed in space. For modeling such data, nonparametric system identification research field is found. Based
on representative points and the weighted distance, this paper discusses nonparametric system identification, and proposes
model estimation method based on classification “consistency” criterion as well. Finally, with typical data such as IRIS and
Breast Cancer data, the effectiveness of model is verified.