Nonparametric Model-Based Estimation In Data Mining

Type Working Paper
Title Nonparametric Model-Based Estimation In Data Mining
Author(s)
Publication (Day/Month/Year) 2010
URL http://stat.upd.edu.ph/docs/research/working papers/2010/Working Paper_2010_13.pdf
Abstract
Probability sampling in finite populations are completely dependent on the
availability of a reliable frame. In market research, especially for new
products/services, the frame that enumerates the target market is not available.
Official statistics like census and survey data are regularly collected by the
Philippine Statistical System. The public use files of these data systems can be
potentially beneficial among researches in the business sector. Using the
weighted household-level data in the 2003 Family Income and Expenditure Survey,
The proposed nonparametric model-based estimation procedure is used to
estimate the market size for food items and some of its components. Model-based
estimation is viewed in the context of re-sampling methods to estimate the
population total. Even if the sample is drawn only from a small part of the
population, model-based estimates are superior or at least comparable to designbased
estimates especially for small populations. In symmetric populations, the
choice of an auxiliary variable (predictor) is important but in a skewed population,
performance of model-based estimator is robust to the relationship between the
target variable and the auxiliary predictor. The bootstrap sampling errors are
generally lower than the design-unbiased sampling errors.

Related studies

»