A novel scheme of classification for non-functional requirements using CNN with LSTM and GRU new hidden layer

In software development processes, finding the requirement before developing the software is essential. There are two kinds of the requirements in software development: functional requirement and Non-Functional Requirement (NFR). For functional requirement a lot of research work has been done but for NFR very limited research has been done. NFR is critical for software development because it specifies quality and constraints of the system. A critical aspect of analysing NFRs is domain knowledge, expertise and significant human effort, since NFRs are written in natural language. To automate the software requirement classification many ML-based techniques are being developed. In this paper, the proposed CNN model obtained the accuracy, recall, precision, and F1-score of 0.984, 0.99, 0.984, 0.984 and 0.989, 0.99, 0.988, 0.999, performance respectively for BOWs and TF-IDF feature selection techniques. The proposed performance varies with respect to the number of requirement classes, but proposed CNN techniques performed better than the existing machine learning techniques.