Robust regression by self-updating process
| Autor: | Chih-Hsuan Wu, 吳芷瑄 |
|---|---|
| Rok vydání: | 2017 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 105 Robust regression based on an M-estimator has been developed by Huber since 1973. A common algorithm is to perform weighted least-squares in which the weights are iteratively updated according to the new fitted line. In this paper, we will present an iterative process reducing the effect from outliers. It is an extension of SUP (self-updating process) clustering algorithm (Chen and Shiu 2007) and mean-shift clustering (Cheng 1995). This process updates the weights and moves the data points to a locally fitted line in each iteration. We also provide some estimation protocols after this process converged. Simulation studies were done to show that our proposed method outperforms the traditional approach in some types of data. For example, (i) data with uniform noise, (ii) heavy-tailed noise data, and (iii) multiple linear models. The convergence problem and some simple examples will be discussed. Finally, a real data set about MLB players’ salaries is analyzed to demonstrate our method. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|