An Ensemble Learning Approach to Software Requirement Classification with Recursive Feature Elimination and Balance Bagging Classifier

Abstract

Software development begins with the critical requirement phase, where user ideas are translated into structured requirements. Clear and accurate documentation of user requirements is fundamental for delivering a high-quality software product. However, this process involves classifying requirements into functional and non-functional categories, which can be time-consuming and error-prone when done manually. To address these challenges, this paper introduces a state-of-the-art model utilizing the Random Forest ensemble learning algorithm for the classification of software requirements. The aim is to enhance the efficiency and accuracy of this classification process. The proposed model was evaluated using precision, recall, F1-score and accuracy performance metrics. The results unveiled the model's exceptional performance, attaining a precision of 0.98, recall of 0.95, F1-score of 0.95, and an astounding accuracy of 0.99, representing a remarkable 99% performance rate. Notably, this performance outshines existing algorithms, specifically the Support Vector Machine (SVM) and K-Nearest Neighbors (KNN), by significant margins, with differences of 0.08, 0.09, 0.09, and 0.24 in precision, recall, F1-score, and accuracy, respectively. In conclusion, the paper recommends the adoption of the Bagging Random Forest model for software requirement classification.