Software Reliability: Development of Software Defect Prediction Models Using Advanced Techniques

Abstract

Development of next generation of smarter machines and services requires building an easy-to-use technology stack. These technologies include hardware, embedded software, data, and applications. Smarter machines support carrying out the precise job, better decision making, and create new ways of doing things to reduce cost and increase speed, accuracy, and automation. As technology meets iron through smart machines, there is increased embedded software within products. To provide distinctive customer experience for a solution system, along with reliability of hardware, development of reliable software also plays a critical role. This paper proposes models in which software reliability is a function of the number of residual faults and is measured with the help of software metrics based on development data. The intent is to establish a statistical relationship between product metrics (that deal with the measurement of the software product) or process metrics (the process by which it is developed) with measures of quality. Using both a pattern recognition algorithm approach for classifying fault proneness and applying fuzzy logic to software metrics for defect prediction is found to be beneficial for improving software reliability prediction in early development phases. This paper proposes non-parametric models like Artificial Neural Network from deep learning to predict expected number of failures utilizing past failure data for software reliability estimation and release readiness during launch.