Comparative Study of an Ensemble Machine Learning Model Versus Maximum Likelihood Model to Assess Reliability Measures in Right Censored Data Analysis

Abstract

This paper explores the estimation of a new power function under Type-II right censoring using two methods: maximum likelihood estimation (MLE) and an ensemble machine learning model based on stacking. The study aims to assess both methods’ effectiveness in estimating various reliability measures, such as hazard rate, mean residual life, variance residual life, mean inactivity time, and variance inactivity time. The stacking model integrates five base models, radial basis function neural network, random forest, Support Vector Regression (SVR), Multilayer Perceptron (MLP), and gradient boosting regression trees, with an radial basis function neural network serving as a meta-learner for final predictions. Numerical experiments compare the performance of the stacking model against MLE for Type-II censored data. Results indicate that the stacking model significantly enhances the accuracy of reliability measure predictions, showcasing its potential as a robust tool for reliability analysis in the context of Type-II censoring