A quantitative Learning Model for Software Test Process Ghaffari Abu Department of Computer Science University of Texas at Dallas Richardson, TX - USA 75083-0688 gabu@utdallas.edu Jo˜ ao W. Cangussu Department of Computer Science University of Texas at Dallas Richardson, TX - USA 75083-0688 cangussu@utdallas.edu Janos Turi Department of Mathematical Sciences University of Texas at Dallas Richardson, TX - USA 75083-0688 turi@utdallas.edu Abstract Learning through experience shows improvement in pro- ductivity. Many models and approaches related to learn- ing or experience curve have been established and success- fully applied to the traditional industries. This paper in- vestigates the learning process and extends the idea of im- provement through learning to software test processes. A novel quantitative learning model for software life cycle is proposed and compared with the existing learning models. An existing formal software test process model is modified to include effects of learning based on the developed learn- ing model. Finally, the extended quantitative software test process model is applied to several industrial software test projects to validate the improved prediction capabilities of the model. 1 Introduction The state variable approach has shown considerable po- tential in modeling of the Software Test Process (STP) 1 [1, 2, 3]. The main advantage of this framework arises from the use of control theory to control, optimize and calibrate the model in a per-company/per-project basis. Formal ap- proaches make the modeling process difficult, however the quantitative results obtained from such models are more ac- curate and practical than the current ad-hoc industrial prac- 1 Software Test Process hereafter will be referred as STP tices. The first level model 2 [3] for STP did not account for important implicit elements affecting the STP. Sensitiv- ity analysis [4] suggests that learning and communication overhead are examples of two such elements. In this paper, an investigation of learning phenomenon during STP is presented and a novel model is proposed to account for learning behavior during software processes. The effect of learning on STP is then analyzed and quan- titatively evaluated by introducing learning parameters into the current state model of STP. As described in Section 6, this analysis shows enhancement in the accuracy of the ex- isting STP model. Incorporation of continuous quantities such as learning and experience into the existing STP model has brought a quantitatively unique and significant improvement in the model. Unlike discrete event simulation models, present model not only goes beyond “what if?” questions but ap- plies closed loop feedback and control mechanism. While formal process languages such as Little-Jil [5] are limited to procedural (coordination in the process steps) aspects of the process, our model at the very least does provide quan- titative estimation of the STP metrics. Present model can be used as complementary model to process models like Lit- tle Jil where the quantitative analysis is not available. The well proven feedback and control approach accompanied by quantitative recommendation to managers is the major strength of this model. Feedback solution in the present model is closed loop and hence provide the opportunity to the project managers to target/achieve the modified qual- 2 Original STP Model over which the current improvement has been applied. 0-7695-2268-8/05/$20.00 (C) 2005 IEEE Proceedings of the 38th Hawaii International Conference on System Sciences - 2005 1