A0847
Title: Local influence on gamma process and trend gamma process
Authors: Yufen Huang - National Cheng Kung University (Taiwan) [presenting]
Abstract: The gamma process (GP) is widely used when the degradation path is strictly increasing. For some circumstances, the GP is not able to successfully describe the degradation path. Hence, random effects are considered in the GP model (REGP) for resolving this problem. Alternatively, a trend gamma process (TGP) has been previously proposed, which integrates the merits of the trend function into a GP model attempting to overcome this obstacle. Case diagnostics play an important role in statistical modelling. For example, during the model fitting process, suspicious observations can greatly influence modelling and forecasting results. Consequently, the detection of such aberrant observations becomes an essential task. To our knowledge, a study on influence analysis for GP, REGP and TGP models has not been explored in the literature. Local influence has been previously proposed to assess the local effect of small perturbations in regression models. Local influence on degradation paths in GP, REGP and TGP models are developed as tools for case diagnostics. Simulation studies and real data examples show the proposed method provides a good tool for outlier/influential path detection as well as an auxiliary diagnosis for assuring the quality of data while using GP, REGP and TGP models.
