Multimodal Generative AI for Story Point Estimation in Software Development

TL;DR

This paper presents a multimodal AI approach combining text, images, and categorical data to improve story point estimation in Agile software development, demonstrating enhanced accuracy and highlighting challenges with complex data categories.

Contribution

It introduces a novel multimodal AI framework for story point estimation, integrating diverse data types to outperform traditional single-modal methods in Agile project management.

Findings

Strong accuracy for simple story points

Challenges with complex categories due to data imbalance

Categorical data like severity significantly impacts performance

Abstract

This research explores the application of Multimodal Generative AI to enhance story point estimation in Agile software development. By integrating text, image, and categorical data using advanced models like BERT, CNN, and XGBoost, our approach surpasses the limitations of traditional single-modal estimation methods. The results demonstrate strong accuracy for simpler story points, while also highlighting challenges in more complex categories due to data imbalance. This study further explores the impact of categorical data, particularly severity, on the estimation process, emphasizing its influence on model performance. Our findings emphasize the transformative potential of multimodal data integration in refining AI-driven project management, paving the way for more precise, adaptable, and domain-specific AI capabilities. Additionally, this work outlines future directions for addressing data variability and enhancing the robustness of AI in Agile methodologies.