Prediction of Sprint Delivery Capability in Agile Software Development Using Machine Learning

Abstract

Scrum plays the dominant framework in agile software development. It delivers the project in an iterative and incremental manner under main concept called sprint. Tracking whether a sprint is capable of delivering its issues on duration is called sprint delivery capability which helps to enhance sprint planning and deliver the software on time. Previous works implemented prediction models using Machine Learning (ML) approach to forecast sprint delivery, but it lacked to capture temporal, hidden patterns and non-linear relationships. To tackle these problems, this research used ensemble learning approach by leveraging the strengths of different ML, Deep Learning (DL) and ensemble algorithms with enhanced feature aggregation and feature selection to fill the gap to improve the prediction of sprint delivery capability in different project progression levels. The data were collected from the JIRA open-source project, consisting of 1873 iterations and 10852 issues in tabular format at 30%, 50%, and 80% project progression levels, was utilized with data preprocessing to clean the data. Feature engineering enabled to improve the performance and interpretability of the model including, feature aggregation and feature selection. The feature aggregation was performed by leveraging statistical and clustering approaches on the features of the issue table to derive new features and merge them with the iteration table. The hybrid feature selection extracted the key features influencing the sprint delivery capability using statistical and ML techniques. The author developed the ensemble models using ensemble methods such as, bagging, boosting and stacking that helped to increase the prediction accuracy. The results demonstrated that 80% progression level provided more accuracy in all the models and Categorical Boosting (CatBoost) algorithm gained highest accuracy of 96.8%. The prediction made on different project progression levels supported to get more insights about sprint performance, identify potential risks and efficiently allocate resources throughout the project lifecycle.