A deep learning framework for non-functional requirement classification.

Area of Science:

• Software Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Classifying nonfunctional requirements (NFRs) from documents is labor-intensive and complex.
• Existing machine learning methods often require time-consuming manual feature extraction.
• Deep learning offers potential for automated feature learning and improved classification accuracy.

Purpose of the Study:

• To propose a novel deep-learning framework for automated NFR classification.
• To overcome the limitations of traditional supervised machine learning in NFR analysis.
• To enhance the efficiency and accuracy of identifying and classifying NFRs.

Main Methods:

• Developed a deep learning framework with a profound architecture for NFR classification.
• Leveraged enhanced representational power and broader context capture compared to shallower models.
• Experimentally evaluated the framework on two established datasets comprising 914 NFR instances.

Main Results:

• The proposed DReqANN model achieved high performance in NFR classification.
• Achieved precision ranging from 81% to 99.8%, recall from 74% to 89%, and F1-score from 83% to 89%.
• Demonstrated superior performance compared to other evaluated models.

Conclusions:

• The deep learning framework is highly effective for NFR classification tasks.
• The DReqANN model shows significant potential for advancing NFR analysis and classification.
• The approach reduces manual effort and improves the accuracy of requirement analysis.