Machine Learning Approach for Testing the Efficiency of Software Reliability Estimators of Weibull Class Models

Usage of software in every field has resulted in the concern over its quality and durability. In this regard, there is a need to have a systematic way of assessing the reliability of the software. One such assessment is the estimation of software reliability. Numerous works have been done in estimating the reliability of the software by making use of software failure times. Most of the software failure times follow Weibull distribution. Herein, Weibull models are considered. Two well-known estimators, viz, the Maximum Likelihood Estimator and the Minimum Variance Unbiased Estimator have been obtained and combined to get Improved Estimator, which satisfies maximum number of statistical properties of a good estimator. In addition, the comparison of the three estimators is carried out by means of coefficient of variation, which considers both the mean and the standard deviation. The comparison is further enhanced by applying statistical tests to these estimators. Machine learning, being the most widely used technique in recent times, herein it is intended to use the R programming language, which is considered as a powerful machine learning language, to carry out statistical tests pertaining to these estimators. Few datasets have been considered and the estimates have been tested for comparison using Modified signed-likelihood ratio test for equality of coefficients of variations. The output results have been analyzed to test the significance of the differences between the coefficients of variation.