|
Probability weighted moments and family of non-parametric regression estimators |
|---|---|
| รหัสดีโอไอ | |
| Creator | Usman Shahzad |
| Title | Probability weighted moments and family of non-parametric regression estimators |
| Publisher | Maejo University |
| Publication Year | 2568 |
| Journal Title | Maejo International Journal of Science and Technology |
| Journal Vol. | 19 |
| Journal No. | 2 |
| Page no. | 160 |
| Keyword | probability weighted moments, non-parametric regression estimators, NW kernel estimator, measures of central tendency, percentage relative efficiency |
| Website title | Maejo International Journal of Science and Technology |
| ISSN | 1905-7873 |
| Abstract | Regression analysis plays a significant role in statistics by identifying the relationship between variables. However, when the assumptions of ordinary least squares are violated, non-parametric regression becomes a preferable approach. In the field of non-parametric regression, Nadaraya-Watson (NW) kernel regression estimators have gained popularity in recent decades. The adaptive NW kernel regression estimator is considered one of the best and most effective estimators for non-parametric regression due to its varying bandwidth. The effective utilisation of the bandwidth factor is a key aspect of kernel regression. This article proposes the use of probability-weighted moments (PWMs) to enhance the bandwidth factor of the kernel regression estimator. The novelty of this approach lies in replacing traditional measures of central tendency and dispersion with PWMs to introduce a new family of NW kernel regression estimators that are more robust to outliers. Monte Carlo simulation studies are conducted to compare the performance of existing and proposed kernel regression estimators. The simulations use a data set of COVID-19 cases from Africa, highlighting the severity of the current pandemic. The results of the simulations demonstrate that the proposed family of estimators is more robust than existing estimators. |
