Software Code Analysis using Ensemble Learning Techniques

Ensuing the advent of advancements in software systems, the probability of them containing high severity defects is exponentially on the rise. With each technological addition, the complexity of software is increasing. Reproduction and rectification of a defect requires time and effort. Current state of the art analysis tools cater to the investigation of static aspects of a production level code. However, it is imperative to assess the dynamic development process of a system so as to be able to timely detect erroneous components early on in the development life cycle of a software. A novel automated defect prediction feature enhancement is proposed that analyses the static structure of the current code and state of the software in past releases to extract relevant static and dynamic feature sets. Data generated is modelled for defect trends in the future release of the software by four ensemble classifiers. Results demonstrate the superiority of Voting algorithm for the problem of defect prediction.