ANALYSIS AND DESIGN OF SOFTWARE RELIABILITY MODEL FOR BIG FAULT DATA USING SOFT COMPUTING

Abstract

With technological advancement in the current scenario, software usage became an integral part of our daily activities ranging from various applications usage through mobile to highly complicated medical devices used in surgeries. Software reliability plays a crucial part in the proper functioning of software at the site and rendering services to the customer. Therefore, it is of utmost importance to eliminate as many faults in the software as possible before its release. The need to deliver a high-quality product becomes a major concern in the industry. Product quality is the only factor that determines its success in the market and can be identified with its reliability. Development of a system became complex and costly due to technological advancement thus one needs to address the criteria regarding security, development cost, and reliability during the development phase to ensure a defect-free, cost effective, and reliable final product. Moreover, a reliability assessment is required on the upgraded versions of the already existing systems. Existing systems are continuously monitored for any possible fault and additional components are added in the new version to address the resulting issues which further required an up-gradation. As the complexity of a system increases so do its functionality and capabilities but since the reliability is inversely proportional to the degree of software complexity achieving a balance between complexity and reliability became difficult. Software companies undergo rigorous testing to remove any probable causes that result in problems and hinder the smooth functioning or reliability of software. Although rigorous testing is carried out to remove software faults they cannot be removed. Developers make use of software reliability models for reliability estimation either by selecting or developing a reliability model suitable for their environmental conditions. The usage of Software reliability growth models (SRGM) for reliability assessment of software quality is centered around the reliability phenomenon. Although reliability models are ideal for measuring and predicting reliability, it's challenging to find an optimal model that works well in all environmental conditions and on different types of datasets. Reliability models are abundant in the literature. Still, not all models can exactly depict reality since while determining the model parameters, there is always iii the possibility of uncertainty. Also, model selection depends on the evaluated parameter's value, comparison criteria for model selection, and fault data set. The parameter evaluation and hence the model’s capability is tied to the usage of a particular data set making predictions less accurate. With the extensive usage of Big-data, the usage of a distributed, high-capacity storage system with a fast-accessing mechanism is required to handle its high speed, high volume, and variety, characteristics which may arise errors in software due to hardware malfunctioning. Similarly, because of unfamiliarity with specific software and an abundant amount of data to handle, give rise to errors in software because of human negligence. Several reliability models have been developed to determine the reliability of a software product assuming that the fault in software results only due to incorrect specifications or errors in code they didn't consider the fault induced in software due to external factors. A great deal of research has been carried out to make use of the various combination of existing models in developing new hybrid models. Parameter evaluation of such hybrid models based on the mathematical equation is very difficult. Their non-linearity and complexity make the statistical parameter evaluation a challenging task. Software reliability models are generally used for the development of such mathematical models which further depends on accurate prediction and parameter optimization based on experimental data. Thus, the motivation of this work is to develop a hybrid model using a combination of NHPP models to handle not only pure software errors but also the errors resulting in software due to hardware malfunctioning and manual intervention without any assumption. This research aims to develop a hybrid model that, apart from pure software errors, also considers induced errors in software due to environmental factors into consideration. A direct modification in an NHPP model that can successfully handle software errors is to combine it with other NHPP models to tackle induced errors resulting from hardware and user. We developed 33 hybrid models by combining NHPP models in various combinations according to their characteristics. To access these developed models, we formulated an estimation function and ranking methodology to select the best model based on the accuracy of estimation. The developed hybrid models were compared with existing traditional models using thirteen comparison criteria. Lastly, soft computing techniques like Genetic Algorithm, Simulated Aneling, and Particle Swarm Optimization were utilized for parameter evaluation and optimization.