A generalized prediction model for improving software reliability using time-series modelling

Abstract

The primary goal of any prediction model is an accurate estimation. Software reliability is one of the software organization's major research priorities. One of the quantitative indicators of software quality is software reliability. The Software Reliability Model is used to assess the reliability at various stages of testing. The purpose of this work is to investigate the software's dependability using time-series modeling, which is the most efficient tool for evaluating its predictive power. A fault prediction model based on categorizing faults for measuring software reliability known as Seasonal-ARIMA (S-ARIMA) is proposed in this work. The significant attribute for complex software applications is to ensure software reliability and fault tolerance. However, these attributes would inculcate additional overheads such as added costs, implementation delay, and the representation of software solution providers. Therefore, the corporation needs to ensure the reliability of the software before delivering it to the clients. Finding the mistake with a decent degree of precision at the right time aims to limit the consequences. We have analyzed and evaluated three real-time data sets to measure software reliability by the proposed prediction model for software reliability. Based on the results of these datasets, the proposed S-ARIMA model has achieved high reliability and improved accuracy when compared with the ARIMA model in terms of different parameters like mean square error ($MSE$), Relative Prediction Accuracy Improvement $\left( { RPAI_{MSE} } \right)$, and Akanke's Information Criteria ($AIC$).