A Systematic Comparison of Software Requirements Classification
| Main Authors: | Baskoro, Fajar; Informatics Department Institut Teknologi Sepuluh Nopember, Andrahsmara, Rasi Aziizah; Informatics Master Programme, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia, Darnoto, Brian Rizqi Paradisiaca; Informatics Master Programme, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia, Tofan, Yoga Ari; Informatics Master Programme, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia |
|---|---|
| Format: | Article info application/pdf eJournal |
| Bahasa: | eng |
| Terbitan: |
IPTEK, LPPM, Institut Teknologi Sepuluh Nopember
, 2021
|
| Subjects: | |
| Online Access: |
https://iptek.its.ac.id/index.php/jts/article/view/13005 https://iptek.its.ac.id/index.php/jts/article/view/13005/6885 |
Daftar Isi:
- Software requirements specification (SRS) is an essential part of software development. SRS has two features: functional requirements (FR) and non-functional requirements (NFR). Functional requirements define the needs that are directly in contact with stakeholders. Non-functional requirements describe how the software provides the means to carry out functional requirements. Non-functional requirements are often mixed with functional requirements. This study compares four primarily used machine learning methods for classifying functional and non-functional requirements. The contribution of our research is to use the PROMISE and SecReq (ePurse) dataset, then classify them by comparing the FastText+SVM, FastText+CNN, SVM, and CNN classification methods. CNN outperformed other methods on both datasets. The accuracy obtained by CNN on the PROMISE dataset is 99% and on the Seqreq dataset is 94%.