Host: The Japan Society for Management Information
There has been a strong need for improving software quality in industry. Quality prediction based on the limited number of software metrics which can be easily collected in practice is highly demanded. The Rayleigh model which can predict the number of total defects based on defect detection history of the target project is one of the promising techniques to meet such needs. However, the model has several problems to be solved for practical application. One of the problems is that the model would predict smaller number of total defects than the number of defects which had already detected. The authors propose a method to resolve this inconsistency by introducing the idea of conditional probability. The effectiveness of the proposed method was validated through real software project data.
