Improving Classification Results Using Class Imbalance Solutions & Evaluating the Generalizability of Rationale Extraction Techniques
Institution: | Miami University |
---|---|
Department: | Computer Science |
Degree: | Master of Computer Science |
Year: | 2015 |
Keywords: | Computer Science; rationale extraction; Chrome Bug Reports; SPSD |
Record ID: | 2059099 |
Full text PDF: | http://rave.ohiolink.edu/etdc/view?acc_num=miami1420335486 |
During the software development process many decisions are made. The decisions, alternatives, and the reasons for and against those alternatives constitute the software design rationale. The research tries to improve the rationale classification results by use of new features. An improvement was observed only for rationale classification but other rationale types showed a F measure reduction. The research also evaluates the generalizability of the machine learning approach using a new dataset though does not establish it. The research implements and experiments with SMOTEBoost algorithm that is used in combination with the new features to give improved results for some categories of rationale. The thesis also describes a tool that can map classified outputs to the original instances that can be used to find missing features.