Optimized Software Reliability Estimation with a Hybrid Support Vector Regression Model

Abstract

Early detection and analysis of software defects helps guide resource allocation for evaluation and verification in software engineering.Software quality assurance involves applying machine learning methods to detect defects, mostly focusing on issues within a single product instead of across multiple products.Reliability prediction models use machine learning to estimate the fault rate in software systems. Many traditional reliability measures are used during debugging and testing to identify software faults.Most machine learning fault prediction models combine with standard  reliability  growth  measures  to  classify  reliability  severity,  often  producing  binary predictions with low standard error.This paper presents a hybrid approach using support vector regression as a non-linear parametric growth measure, applied to training fault datasets. Results are tested on several reliability datasets with different parameter settings for fault prediction.