EstiMate: Bridging the Gap between Agile Story Point Prediction and Project Management Platforms

Abstract

"Effort estimation plays a major role in software development projects. When it comes to the agile estimation process, story point estimation plays an important role in defining the weights of product backlog tasks. Within recent years, several studies have been conducted on developing Machine Learning models to predict effort estimation in story points. With the advancement of Natural language processing techniques, some models such as GPT2SP , DEEP-SE and SE3M, have the capability to capture the contextual information of the given tasks description and make predictions accordingly. However, these were not applied to real-world usages so far; most of them were limited to a Python notebook or a simple demonstration application. Meanwhile, most software development companies execute their agile workflows using project management platforms and these platforms lack the ability to provide estimation predictions or early insights about the development tasks. EstiMate is a novel integration mechanism which can be used to fill the gap between mentioned prediction models and project management platforms (PMPs).

The proposed system utilises both task titles and descriptions in the updated variant of the GPT2SP model with a novel data clearing algorithm specifically designed to filter out the noisy syntaxes added by the project management tools. Integration of task titles and descriptions was experimented with in two ways: passing the concatenated title and description to the transformer (concatenated approach) and a combination of two GPT2 models for title and description (ensemble approach). Furthermore, the system integrates itself into the JIRA platform as a plugin, effectively bringing the story point prediction ability into project management tools.

Several experiments were carried out to incorporate the task descriptions into the GPT2SP model. And results have shown that using the concatenated approach was more accurate than using ensemble models. Also, the evaluations obtained from the problem domain experts showed that the implementation of the effort estimation model directly into the project management platforms was highly beneficial."