A Machine Learning Ensemble–Based Data-Driven Model for Software Reliability Growth

Software reliability growth modelling is an essential aspect of evaluating the quality and dependability of software systems throughout both testing and operational phases [1,2]. Conventional and statistical Software Reliability Growth Models (SRGMs), while well established in theory, often face limitations when dealing with the complex, nonlinear, and data-driven behaviour observed in real software failure patterns. To address these challenges, this study proposes a purely data-driven software reliability growth model using machine learning ensemble methods. The proposed framework combines bagging- and boosting-based algorithms, including Random Forest, AdaBoost, and Gradient Boosting, to effectively learn failure trends from historical inter-failure time and cumulative failure data. Comprehensive experiments are performed using widely used public software reliability datasets. The experimental results indicate that ensemble-based models consistently achieve higher prediction accuracy and better generalisation performance compared to traditional machine learning approaches and classical regression-based models. These findings highlight the strength of ensemble learning in capturing complex reliability growth behaviours, especially when dealing with limited or noisy datasets. Overall, the proposed model will provide a reliable and scalable solution for software reliability assessment and support informed decision-making in practical software quality assurance environments.