A Fine Tuned Ensemble Approach to Classify Requirement from User Story

The manual classification process in software requirement Specification (SRS) has several challenges, including subjectivity, time commitment, and the possibility of mistakes and discrepancies. There is a chance of inconsistencies and classification errors when requirements are manually classified because of subjectivity and anomalies. Furthermore, the corresponding manual effort grows increasingly difficult and unfeasible as the volume of requirements increases. This study explores the field of automated classification, highlighting the crucial role of automation in distinguishing between functional and non-functional requirements. Utilizing the PROMISE repository with 969 requirements, the study deploys an ensemble of machine learning classifiers—Random Forest, Multinomial Naive Bayes, Gradient Boosting, XGBoost, and AdaBoost. Based on the test dataset, the empirical results show that the suggested ensemble method achieves a 98% accuracy in binary classification of Functional Requirements (FR) and Non-Functional Requirements (NFR). Moreover, Grid search and hyperparameter optimization methods refine the ensemble’s performance, highlighting its usefulness in enhancing the accuracy and consistency of software requirement analysis in the binary classification framework.