Abstract — During software development two important decisions organizations have to make are: how to allocate testing resources optimally and when the software is ready for release. SRGMs (Software Reliability Growth Models) provide empirical basis for evaluating and predicting reliability of software systems. When using SRGMs for the purpose of optimizing testing resource allocation, the model’s ability to accurately predict the expected defect inflow profile is useful. For assessing release readiness, the asymptote accuracy is the most important attribute. Although more than hundred models for software reliability have been proposed and evaluated over time, there exists no clear guide on which models should be used for a given software development process or for a given industrial domain. Using defect inflow profiles from large software projects from Ericsson, Volvo Car Corporation and Saab, we evaluate commonly used SRGMs for their ability to provide empirical basis for making these decisions. We also demonstrate that using defect intensity growth rate from earlier projects increases the accuracy of the predictions. Our results show that Logistic and Gompertz models are the most accurate models; we further observe that classifying a given project based on its expected shape of defect inflow help to select the most appropriate model. Keywords — Embedded Software; SRGMs; Defect Inflow; Software reliability growth models; Automotive; Telecom; Defense Industry I. INTRODUCTION EMBEDDED software is today an integral part of most products, on which we depend for smooth functioning of our daily life. Embedded software does not only provide functionality, it also drives innovation in mobile phones, satellite systems, home appliances, and aircrafts. Reliability is an important attribute of such systems and one way of evaluating their reliability is to use Software Reliability Growth Models (SRGMs). SRGMs are the result of applying reliability engineering theory to the software development domain. The defect inflow data is modelled using mathematical models that quantify the change in reliability of the given software artefact during its development and testing. SRGMs help to answer an important practical question as to when the given software quality is good enough and thus, when can we stop testing (Dalal and Mallows, 1988). The good- enough quality is also referred to as release readiness of a given product (Kapur and Bhalla, 1992). From the reliability standpoint, one of the most important factors for deciding if a software is ready for release is the number of remaining defects (latent defects). By comparing the predicted total number of defects (asymptote of SRGMs) and the number of defects discovered and resolved to date, software managers can decide if the software is ready to be released (Quah, 2009). Apart from answering the important release readiness question, SRGMs can also be used to make the software testing process more efficient (Malaiya et al., 1992). However, requirements for the successful application of SRGMs for optimal resource allocation and the assessment of release readiness of software differ. Models which can be applied early in the project and have higher ability to accurately forecast the expected shape of the defect inflow profile are useful for optimizing test resource allocations. While SRGMs that are accurate in forecasting total expected defects in a software product (asymptote) late in the development/testing phase are better suited for assessing the release readiness of a given software system. Although more than hundred SRGMs have been proposed and evaluated in the literature (Lyu, 2007), many of the earlier studies evaluating SRGMs have focused only on how well they could fit to the observed defect inflow data. The evaluation of the predictive power of SRGMs in the literature has generally been limited to only the last few data points (typically last 10% of data) (Rana et al., 2013a) (Pham, 2003). The difficulty of applying Selecting software reliability growth models and improving their predictive accuracy using historical projects data Rakesh Rana 1* , Miroslaw Staron 1 , Christian Berger 1 , Jörgen Hansson 1 , Martin Nilsson 2 , Fredrik Törner 2 , Wilhelm Meding 3 and Christoffer Höglund 4 1 Computer Science & Engineering, Chalmers/ University of Gothenburg, Göteborg, Sweden 2 Volvo Car Corporation, Göteborg, Sweden 3 Ericsson, Göteborg, Sweden 4 SAAB AB, Göteborg, Sweden