A Novel Deep Learning Model for Effective Story Point Estimation in Agile Software Development
Harish Mittal, Mohd Arsalan, Puneet Garg
Abstract
Effort estimation is crucial for successful software project management, with accuracy being pivotal for planning and monitoring. While traditional projects have seen extensive research in this area, agile development, particularly in estimating user stories and critical issues, lacks sufficient exploration. To address this gap, our paper introduces a tailored dataset for story points-based estimation, comprising 23,313 issues from 16 open-source projects. We propose a novel deep learning model, the Long-Deep Recurrent Neural Network (LD-RNN), combining LSTM and recurrent highway network (RHN) architectures. This end-to-end trainable system outperforms existing baselines and alternatives, enhancing accuracy in agile contexts. Our article provides a detailed analysis, covering dataset creation, LD-RNN model intricacies, experimental results, and implications for agile development. The findings contribute to the evolving landscape of agile software development, improving effort estimation through innovative deep learning methodologies. The LD-RNN model offers a promising avenue for more precise project planning and resource allocation in agile environments, addressing the unique challenges posed by user stories and issues.