Non-Functional Requirements Classification with Feature Extraction and Machine Learning: An Empirical Study

Abstract

Non-Functional Requirements (NFR) describe a set of quality attributes required for a software such as security, reliability, performance, etc. Extracting and considering NFR from software requirement specification can help developers to deliver quality software which meets users expectations completely. Since, the functional and non-functional requirements are mixed together within the same SRS, it requires a lot of human effort for distinguishing them. This paper proposed automatic NFR classification approach for quality software development by combining machine learning feature extraction and classification techniques. An empirical study with seven machine learning algorithms and four feature selection approaches have been applied to automatically classify NFR for finding out the best pair. The experiments were measured with statistical analysis including precision, recall, F1-score, and accuracy of the classification results through all the combinations of the techniques and algorithms. It is found that, SGD SVM classifier achieves best results where precision, recall, F1-score, and accuracy reported as 0.66, 0.61, 0.61, and 0.76 respectively. Additionally, TF-IDF (character level) feature extraction technique illustrated higher average score than others.