Finding error-prone classes at design time using class based Object-Oriented metrics threshold through statistical method

Abstract

A study that how error severity categories depend on the class level software metrics is presented through statistical method. The main purpose of the study is to classify error categories based on the different number of error occurrences in all the three version of Eclipse Project. The study used the all error type to find the software metrics threshold for the three releases of Eclipse project using Receiver Operating Characteristic curves. These thresholds are responsible for making difference between error-free or error prone classes . But, not all the choosen metrics are able to do that, though some of them are capable for that. In future it is not necessary that these software metric thresholds can predict the class will definitely have errors. This approach only provide a scientific way for software engineers to judge designed class is error prone or error free during design time.