A Machine Learning based Traceability Links Classification: A Preliminary Investigation

Traceability link recovery (TLR) is an effective tool for software engineers to better understand the connection between the high-level and low-level artifacts found in most projects. Most research papers published in the area leverage information retrieval techniques and formulate the TLR activity as a retrieval problem as it provides the user with a collection of possible links that they can go through and validate related documents. However, it still requires significant amount of human involvment which can slow down the tracing process. In this research, we address this problem by transforming it into a simple binary classification problem. The paper presents what features help benefit the overall process of classifying the possible links as well as the classification algorithms used. The results show that Random Forest outperforms the other four classification techniques.