On an Efficient Algorithm for Boundary Estimation

10.29973/JCSA.200712.0003

Tsung-Lin Cheng；Xue-Wen Lu

Change-curve detection ； random fields ； boundary estimation

中國統計學報

45卷4期（2007 / 12 / 01）

372 - 385

英文

We propose a novel and efficient algorithm to detect the change-curve which separates a random field into two parts with distinct distributions. This algorithm, compared with the cumulative-sums (CUSUM) estimations in the one-dimensional case (cf. Cheng 2006), not only simplifies the proofs of the consistency of the estimators of the change-point but also dramatically reduces the computation time without sacrificing too much precision. We extend this method to the random fields under mild conditions. To illustrate the algorithm, we perform some Monte Carlo simulations for different models with different changes.

1. Carlstein, E.,Krishnamoorthy, C.(1992).Boundary estimation.Journal of American Statistical Association,87,430-438.
2. Cheng, T.L.(2006).`An efficient algorithm for Estimating a Change-Point`.under revision for Statistics & Probability Letters
3. Krishnamoorthy, C,Carlstein, E.(1994).Practical considerations in boundary estimation: model-robustness, efficient computation, and bootstrapping.Changepoint Problems, IMS Lecture Notes-Monograph Series,23,177-193.
4. Puri, M.L.,Ruymgaart, F.H.(1994).Change curves in the presence of dependent noise.Change-point Problems, IMS Lecture Notes-Monograph Series,23,242-254.