Adaptive Learning for Transparent NFR Classification: Boosting Software Quality with XAI and Model Efficiency

This paper present a framework for identifying and classifying non-functional requirements (NFRs) by leveraging advanced unsupervised and semi-supervised learning approaches. It addresses the significant reliance on extensive labeled datasets and domain-specific biases inherent in traditional methods. Through the integration of transfer learning and self-supervised strategies, the framework enhances model efficiency and generalizability across diverse contexts. Crucially, the incorporation of Explainable AI (XAI) methods ensures transparency and interpretability, fostering a crucial feedback loop between practitioners and automated systems for continuous refinement. This theoretical exploration along with validation paves the way for more adaptive and accurate NFR analysis, ultimately contributing to improved software quality and development processes.