Machine Learning Based Software Defect Categorization Using Crowd Labeling

Defect categorization is an important task which helps in software maintenance. It also helps in prioritizing the defects, resource allocation, etc. Standard machine learning techniques can be used to automate the categorization of defects. Labeled data is needed for learning models. The expert is required for obtaining the labeled data. Sometimes, it is costly or expert is not available. So, to overcome this dependency, crowd labeled data is used to train a model. Crowd (a set of novices) is asked to assign a category as defined by IBM’s Orthogonal Defect Classification (ODC) to the defect reports. Obtaining categories through crowd can be inaccurate or noisy. Inferencing ground truth is a challenge in crowd labeling. Support Vector Machine, k Nearest Neighbor and Gaussian Naive Bayes classifier, are learnt effectively using new methodology from data labeled by a set of novices. In this chapter, we have proposed a learning model which learns effectively to predict the impact category of software defects using the expectation maximization algorithm and shows the better performance according to the various types of metrics by improving the existing technique by 8% and 11% accuracy for Compendium and Mozilla datasets respectively.