Defect prediction for embedded software

Abstract

As ubiquitous computing becomes the reality of our lives, the demand for high quality embedded software in shortened intervals increases. In order to cope with this pressure, software developers need new approaches to manage the development cycle: to finish on time, within budget and with no defects. Software defect prediction is one area that has to be focused to lower the cost of testing as well as to improve the quality of the end product. This research proposes a defect prediction model specifically for embedded software systems. We utilize machine learning techniques in order to identify the complex relationship between software metrics and defect-proneness. Our proposed model involves three different machine learning techniques which have useful characteristics for defect prediction in embedded software. We combine the strengths of these machine learning techniques in order to obtain a general model for defect prediction in embedded software. The resulting model may be used to assist the embedded software developers in planning the future iterations of their development life-cycle and allocating their limited testing resources more effectively. This will help embedded software developers in increasing the quality of their products by increasing the efficiency of defect removal strategies.