Reliability modelling using ranking algorithm for parameter evaluation

Big data delivers high velocity, high volume, and high veracity data in vast quantities. Analyzing and handling this data requires new software, fast and high-capacity hardware, and trained individuals. One can determine the quality of such data models by assessing the software's reliability at a certain level after considering all the external and environmental factors affecting the software functionality. There are many existing reliability models, but the performance of these models regarding Big Data is still questionable. Thus, we need to develop a model that best suits the big data environment and accurately predicts the fault likely to occur. We developed a hybrid model to determine the software reliability using big fault data. The model included external factors, like human interaction and hardware malfunctioning resulting from the volume, veracity, and velocity of Big Data while determining software reliability, giving better performance and estimation accuracy than the existing well-known models. This paper proposes ten new hybrid models used for handling software reliability of big data. The intensity function and mean value of the proposed hybrid models were derived by combining well-known models' mean value and intensity value function. The parameter evaluation was done using the maximum likelihood method and ranking algorithm within the specified range of the guess vector. The data was collected using Big Data analysis. The best model out of the proposed ten is selected based on estimation accuracy and weighted function value. The chosen model was validated using a set of comparison criteria and estimation accuracy comparison. A t-test was also done to show that both the developed model's estimated and observed values were drawn from the same population pool. The comparisons and experiment results strengthen our claim regarding the better performance of our developed model over existing ones.