General EM algorithm for fitting non-monotone hazard functions from truncated and censored observations |
| |
Institution: | 1. Innovation Center, Samsung Electronics, Hwasung, Gyeonggi-do, 18448, Republic of Korea;2. Department of Industrial & Systems Engineering, Korea Advanced Institute of Science and Technology, 291 Daehak-ro, Yuseong-gu, Daejeon 34141, Republic of Korea;3. Department of Industrial and Management Engineering, Pohang University of Science and Technology, Pohang, Gyeongbuk, 37673, Republic of Korea |
| |
Abstract: | Recently, many researchers focused on modeling non-monotonic hazard functions such as bath-tube and hump shapes. However, most of their estimation methods are focused on complete observations. Since reliability data are typically censored and truncated, a general EM algorithm is proposed, which can fit any of those complex hazard functions. The proposed EM algorithm is analyzed by fitting well-known 4-parameter hazard functions, where its performance is compared by their specific direct methods through extensive Monte Carlo simulations. |
| |
Keywords: | EM algorithm Sequential quadratic programming Phase-type distribution Left-truncated & right-censored Hazard function |
本文献已被 ScienceDirect 等数据库收录! |
|