Non Functional Requirements Identification and Classification Using Transfer Learning Model

Abstract

In this research study, we address the critical task of identifying and classifying non-functional requirements (NFRs) in software development. NFRs, described in the software requirements specification (SRS) document, offer a comprehensive system view and are closely aligned with software design and architecture. However, they are often overlooked compared to functional requirements, leading to potential issues such as rework, increased maintenance efforts, and inefficient resource utilization, impacting project cost and budget. To streamline software development, we propose a novel approach based on transfer learning methods to automate NFR identification and classification, aiming to reduce development time and resource consumption, ultimately leading to improved efficiency. We evaluate multiple state-of-the-art transfer learning models, including XLNet, BERT, Distil BERT, Distil Roberta, Electra-base, and Electra-small, for this purpose. Among them, XLNet demonstrates exceptional performance, achieving an impressive value of 0.91489 for Accuracy, Precision, Recall, and F1 Score. This research highlights the importance of considering non-functional requirements (NFRs) in software development and the negative consequences of neglecting them. It also emphasizes the benefits of using the XLNet tool to automate the identification and classification of NFRs. By using XLNet, we aim to make software development easier, optimize resource usage, and improve the overall quality of software systems.

Keywords

• Transfer learning
• non-functional requirement
• XLNet
• classification and SRS