PERBANDINGAN LIMA METODE FITTING MODEL REGRESI WEIBULL DENGAN PENDEKATAN BAYESIAN:

STUDI KASUS PADA DURASI MENDAPATKAN PEKERJAAN KEMBALI SAAT PANDEMI COVID-19 DI KALIMANTAN TIMUR

Abstract

The Weibull distribution be not included in the exponential family, so this distribution has no canonical link function. There are several parameterizations of the Weibull distribution and the link function used in the regression model. The difference in parameterization and link function makes the method of fitting the model also different. In this study, the five adjustment methods used in the Weibull regression model are described. The model was applied to determine the duration of returning to work during the COVID-19 pandemic in East Kalimantan. The estimation of the parameters of the five fitting methods is carried out using the Bayesian approach of the Hamiltonian Monte Carlo (HMC) algorithm. The results showed that the Weibull regression model fitting method using the natural logarithmic transformation was the best, as indicated by the minimum cross-validation criteria Pareto-Smoothed Important Sampling Leave-One-Out (PSIS-LOO ).