The Effect of Environmental Metrics on Software Fault Prediction

Abstract

In this study, besides the software metrics, the environmental metrics such as experience of software engineer, similar project experience, size of the project, programming language, time spent on analysis and development are also explored to see whether they also affect the results of software fault prediction and what would be the success rates. The dataset for this study was generated from combining various data from 10 projects. A total of 36 metrics and 6676 test cases were evaluated. The errors occurred in the test cases are not just considered as an error, their priority and cases that cannot be tested are also taken into consideration. Nine fault levels are employed in models. Models are created with four different algorithms which have achieved a success rate of; 76% by the decision tree algorithm, 94% by the nearest neighbors algorithm, 90% by the random forests algorithm and 73% by the Adaboost Classifier Algorithm. It was observed that environmental metrics are indeed effective in software fault prediction and when applied with machine learning algorithms a high rate of success can be achieved.