Software Reliability Growth Modeling Based on Generalized Lindley Distribution
Shiv Kumar Sharma, Abhishek Thakur · 2024
In the realm of software development, where the stakes are high and the demand for reliability is paramount, the efficacy of Software Reliability Growth Models (SRGMs) becomes crucial. This study focuses on comparing the effectiveness of a Generalized Lindley Distribution (GLD) model against the traditional Exponential and Weibull models in predicting software reliability. Utilizing hypothetical failure time data from a software system, the models are evaluated across several metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Log-Likelihood, Akaike Information Criterion (AIC), R-Squared (R²), and the Kolmogorov-Smirnov test. The investigation is further deepened through a 5-fold cross-validation approach, ensuring the robustness and stability of the models across different data subsets. The GLD model consistently outperforms its counterparts in terms of lower MSE, RMSE, and AIC values, demonstrating a superior fit to the data and an optimal balance between model complexity and accuracy. These results highlight the potential of the GLD model as a more effective tool in software reliability assessment, crucial for the timely and cost-effective release of highly reliable software in today's technology-driven world.