ÁP DỤNG MÁY HỌC VÀO ƯỚC LƯỢNG CÔNG SỨC PHÁT TRIỂN TRONG PHƯƠNG PHÁP PHÁT TRIỂN PHẦN MỀM LINH HOẠT

Tóm tắt

Software effort estimation in Agile Software Development remains challenging because traditional approaches such as Planning Poker and expert judgment are often subjective and inconsistent. This study investigates the application of Machine Learning techniques for Story Point estimation using a large-scale public dataset containing 23,313 issue reports collected from 16 open-source Agile projects. Textual information from issue titles and descriptions is processed using TF-IDF, Word2Vec, and Doc2Vec feature extraction techniques before being evaluated with several regression-based models, including Random Forest, Support Vector Regression, Gradient Boosting, XGBoost, and Long Short-Term Memory networks. In addition to reproducing previous Story Point estimation approaches, this study proposes and compares Support Vector Regression and XGBoost Regressor models to improve prediction performance. Experimental results demonstrate that Machine Learning-based approaches can effectively reduce estimation errors and improve the consistency of Agile software effort estimation on large-scale datasets.