Applications of Non-Parametric Density Estimation
Abstract
The dissertation examines various methods of nonparametric density estimation, and nonparametric kernel estimation in more detail. The consequences of various kernel window width and their effect on the mean integrated square error are examined using Monte Carlo techniques. The mean and the variance of nonparametric density estimator is derived for symmetric kernels with finite mean and finite variance. The results also treat kernels with varying window parameters. The nonparametric kernel estimate was used to obtain new estimators for the three parameter Weibull distribution using distance estimation and the Cramer-von-Mises statistic. Comparison with maximum likelihood estimators using a Monte Carlo sample of size 1000 and various different parameters showed a significant improvement over the maximum likelihood estimators in the mean integrated square error between the estimated distribution and the true distribution. Several new goodness of fit tests are proposed using the nonparametric kernel estimator and the Cramer-von-Mises and the Anderson Darling statistics. Extensive Monte Carlo experiments were performed to obtain the critical values for the test and to study the power of the tests against eight alternative distributions. The tests using the Anderson Darling statistic showed greater power against almost all alternative distributions studied than the K.S. test.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 28, 1990
- Accession Number
- ADA220206
Entities
People
- Ahmed M. Sultan
Organizations
- Air Force Institute of Technology