Predicting type II censored data from factorial experiments using modified maximum likelihood predictor

Abstract

Censored data are often found in industrial experiments. The censored data are usually predicted by constructing complex statistical models or neural networks. Although a maximum likelihood predictor (MLP) was developed to predict Type II censored data, the likelihood equation may not be obtained for a closed-form solution. A modified maximum likelihood predictor (MMLP) was derived to overcome the problems of MLP. However, because MMLP requires normality assumption with unknown mean and known variance, and because the population variance of real-world experimental data is generally unknown, the MMLP has little practical use. Therefore, this study develops a modified maximum likelihood predictor (MMLP) for Type II censored data obtained from a normal distribution with unknown mean and variance. The predicted censored data using the proposed MMLP are merged with the uncensored data as a pseudo-complete data set. The analysis of variance (ANOVA) method is then employed to determine the optimal factor-level combination settings. The proposed method can also be employed to predict the Type II censored data obtained from Taguchi's parameter designs. Two examples are given to demonstrate the proposed method and the comparisons of the proposed method with existing methods of predicting the Type II censored data are made to demonstrate the effectiveness of the proposed method.