Investigating the Effectiveness of Clustering for Story Point Estimation

Abstract

Automated techniques to estimate Story Points (SP) for user stories in agile software development came to the fore a decade ago. Yet, the state-of-the-art estimation techniques' accuracy has room for improvement.In this paper, we present a new approach for SP estimation, based on analysing textual features of software issues by employing latent Dirichlet allocation (LDA) and clustering. We first use LDA to represent issue reports in a new space of generated topics. We then use hierarchical clustering to agglomerate issues into clusters based on their topic similarities. Next, we build estimation models using the issues in each cluster. Then, we find the closest cluster to the new coming issue and use the model from that cluster to estimate the SP.Our approach is evaluated on a dataset of 26 open source projects with a total of 31,960 issues and compared against both baselines and state-of-the-art SP estimation techniques.The results show that the estimation performance of our proposed approach is as good as the state-of-the-art. However, none of these approaches is statistically significantly better than more naive estimators in all cases, which does not justify their additional complexity. We therefore encourage future work to develop alternative strategies for story points estimation.The experimental data and scripts we used in this work are publicly available to allow for replication and extension.

A SLR on Deep Learning Models Based on Textual Information For Effort Estimation in Scrum

“…This has led to numerous studies over the years to help with effort estimation and improve the accuracy of effort estimation using Machine Learning algorithms. Several of these studies have been trying to uncover automated ways of story points effort estimation for tasks based on certain features with the intention of averting incorrect effort estimations due to subjectivity and bias resulting from human judgement, along with, and more essentially to an agile team, yielding consistent estimates across different sprints and releases and the overall project's lifecycle [3].However, there are no reviews which focus of the use of Deep Learning algorithms with special focus on story points effort estimation. The unavailability of appropriate information, metrics as well as past experiences around work item effort estimation can surely lead to imprecision issues.…”

“…Agile teams generally rely on using Story Points as the unit of measure [3], for estimating the combined effort required by each user story, thereby relying on expert-based estimation, with additional buffers for any uncertainty or added complexity within those user stories [4]. As opposed to traditional approaches which aimed at estimating the whole project, estimation using story points gauges the required effort for each individual user story or feature.…”

Story points changes in agile iterative development

“…To evaluate the accuracy of the models, we measure the Standardized Accuracy (SA). In particular, SA measures how well our models can estimate the development time in comparison to random guessing (Sarro et al 2016;Tawosi et al 2021)…”

“…Latent Dirichlet Allocation (LDA)-based classifier: A recent study showed that LDAbased classifiers can be used to predict the SP (Tawosi et al 2022). Hence, we used LDA-based classifier as our baseline to evaluate whether our classifier can outperform a classifier built based on the topics extracted from natural language text.…”

“…Since effort estimation is time-consuming (Hoda and Murugesan 2016), prior studies proposed automated approaches to predict the effort at the project level (Manga and Blamah 2014;Kuan 2017;Sarro and Petrozziello 2018;Sarro et al 2016;Tawosi et al 2021) and the sprint level Ramessur and Nagowah 2021). Recent approaches were proposed to predict the effort at the work item level (Porru et al 2016;Scott and Pfahl 2018;Choetkiertikul et al 2019;Alhamed and Storer 2021;Fu and Tantithamthavorn 2022;Tawosi et al 2022). Although these approaches could help practitioners estimate and update the effort anytime, the change of effort estimation after sprint planning could invalidate the sprint plan.…”