A Multi-Model Machine Learning Framework for Automated Requirement Prioritization in Software Development

Abstract

Software requirement prioritization is a critical decision-making activity in agile software development, directly influencing project success, resource allocation, and stakeholder satisfaction. Traditional prioritization methods, such as the Analytic Hierarchy Process (AHP) and MoSCoW, rely heavily on subjective human judgment and become cognitively overwhelming as project scale increases. Recent advancements, such as the framework proposed by Jamasb et al. (2025), have leveraged Natural Language Processing (NLP) to automate requirements management. Their pipeline uses BERT embeddings to extract semantic meaning, K-Means clustering to group similar requirements, and similarity computation to generate a structured requirements list. However, their approach remains fundamentally unsupervised; it organizes and groups requirements for human review but does not autonomously predict a final priority score. This paper proposes a fully automated, supervised machine learning (ML) framework that eliminates the need for human intervention entirely. We evaluate four regression algorithms—XGBoost, Random Forest, Support Vector Machine (SVM), and Artificial Neural Networks (ANN)—on the publicly available UCL, Smart Card requirements dataset, utilizing stakeholder roles, influence weights, and semantic text features extracted via TF-IDF. Our experimental results demonstrate that the XGBoost model significantly outperforms the Jamasb et al. (2025) baseline in Mean Absolute Error (MAE), achieving a score of 0.0678 compared to their reported 0.9000. Furthermore, our model achieves a competitive F1-Score of 0.4560 against their 0.6000, while operating entirely without manual expert intervention. Additionally, the XGBoost model achieved a Spearman Rank Correlation of 0.4156, proving its effectiveness in preserving the correct order of importance. This work establishes that supervised ML pipelines provide a superior, scalable, objective, and cost-effective alternative for modern agile requirement engineering.