Automatically Classifying Non-functional Requirements with Feature Extraction and Supervised Machine Learning Techniques: A Research Preview

Abstract

Context and Motivation: In large projects, extracting the relevant NFR-information as per the stakeholder’s responsibility and needs can be time-consuming and challenging. Question/Problem: Classification of NFRs is one way to mitigate this problem. However, because of the size and complexity of the SRS, the manual classification of NFRs is considered time-consuming, labour-intensive, and error-prone. An automated solution is needed that provides a reliable and efficient classification of NFRs. Principal ideas/results: Using natural language processing and supervised machine learning (SML) algorithms, we investigate feature extraction techniques (i.e., POS-tagging based, BoW, and TF-IDF) to assess their efficacy in automated classification, in conjunction with the SML algorithms (such as: SVM, SGD SVM, LR, DT, Bagging DT, Extra Tree, RF, GNB, MNB, and BNB). Contribution: The proposed combinations: (i) SVM with TF-IDF, (ii) LR with POS and BoW, and (iii) MNB with BoW, all achieve precision and recall values greater than 0.85, and process execution time of less than 0.1 s. Comparison with related work is favourable as is preliminary validation using an industry dataset.