Requirement Classification using Deep Learning and Nature-Inspired Optimization Technique
Sonal Sonawane, Shubha Puthran · Informatica · 2025
Recent advancements in requirements engineering (RE) methods have increasingly leveraged machine learning (ML) algorithms to address challenging RE issues such as the identification and classification of software requirements in requirement documents. Classifying software requirements accurately and efficiently is crucial for the success of the project. Functional and nonfunctional requirements define the core attributes and constraints of a system. When requirements are documented in natural language, they often introduce uncertainties such as ambiguity, inconsistency, or poor readability making the process challenging. Further, manual extraction of requirements is tideous and error prone as it requires precise interpretation, thereby mitigating the risks of miscommunication and errors in the development process.To address these challenges, it is vital to employ Natural Language Processing (NLP)techniques to enhance the clarity of requirements. This paper proposes a method that combines deep learning models with NLP techniques, supplemented by Flower Pollination Optimizer (FPO) algorithm, to automate the classification of requirements. The methodology leverages NLP to extract significant features for training a Convolutional Neural Network (CNN) model. CNN model is enhanced using the FPO algorithm to ensure better convergence. The performance of the proposed CNN-FPO framework is evaluated using an industry SmartNet dataset, with results indicating that the accuracy of the classification can reach up to 98% as compared to other traditional machine learning(ML) approaches.The integration of NLP with an optimized CNN model creates a robust framework for requirement classification, addressing the challenges posed by natural language documentation in requirements engineering.