Pre-trained Model-based NFR Classification: Overcoming Limited Data Challenges

Machine learning techniques have shown promising results in classifying NFR. However, the lack of annotated training data domain requirement engineering poses challenges to accuracy, generalization, and reliability ML-based methods, including overfitting, poor performance, biased models, out-of-vocabulary issues. This study presents an approach for classification non-functional requirements (NFRs) software specification documents by extracting features from word embedding pre-trained models. The novel algorithms are specifically designed extract relevant representative In addition, each model is paired with four tailored neural network architectures NFR RPCNN, RPBiLSTM, RPLSTM, RPANN. combination creation twelve unique its configuration characteristics. show that integration GloVe models RPBiLSTM demonstrates highest achieving impressive average Area Under Curve (AUC) score 96%, a precision 85%, recall 82%, highlighting strong ability accurately classify NFRs. Furthermore, among Word2Vec RPLSTM achieved notable results, AUC 95%, 86%, 80%. Similarly, integrated fastText-based yield competitive comprehensive provides practical solution effectively analyzing NFRs, thereby facilitating improved development practices.