Requirements Classification for Requirements Reuse

In various domains, standards are used to ensure a high level of product quality. During standard tailoring, requirements from the applicable standards are specialized and integrated into the project. The requirement type influences the way the standard requirement interacts with project requirements. Yet, manual classification of large existing standards is time-consuming. This thesis presents a machine learning pipeline to compare four algorithms for this task: k-Nearest Neighbor (kNN), Support Vector Machine (SVM), Logistic Regression (LR), Multinomial Naive Bayes (MNB), as well as an ensemble model combining all four. The models are trained and tested with 466 requirements from the European Cooperation for Space Standardization (ECSS). SVM and LR achieve the best results with F1 scores around 0.85. The integration of term contexts could potentially further increase the prediction accuracy. Yet, the improvement for our dataset is insignificant.