A naive Bayesian Belief Network model for predicting effort deviation rate in software testing

Most of projects´ cost exceeds 10% of yearly corporations´ turnover, a major factor contributing to this loss is the overrun cost of software testing. A lot of events during software Quality Assurance(QA) cycles, the main execution part of testing process, lead the loss. Therefore, there is a great potential benefit to find a way to predict the loss when the risk events arise or when we know they will happen during the QA cycles. In this paper, a model is proposed to solve the above problem via Bayesian Belief Network (BBN), in this model, five independent factors, which may lead the loss in software testing, are extracted by exploring the historical documents and questionnaires back from QA managers, and they are used to classify the loss of the QA effort, and predict the probability distribution of the loss, the mean of the distribution is defined as the predicated loss. The model is proved effective according to the data collected from 45 delayed QA cycles.