Data Science Methodologies: Current Challenges and Future Approaches

TL;DR

This paper reviews existing data science methodologies, highlights current organizational and technical challenges, and proposes a holistic framework to improve project management and execution in data science.

Contribution

It provides a comprehensive review of current methodologies and introduces a new conceptual framework for holistic data science project management.

Findings

Existing methodologies are outdated and focus mainly on technical issues.

Many methodologies lack guidance on team and data management.

The proposed framework aims to address organizational and technical challenges.

Abstract

Data science has employed great research efforts in developing advanced analytics, improving data models and cultivating new algorithms. However, not many authors have come across the organizational and socio-technical challenges that arise when executing a data science project: lack of vision and clear objectives, a biased emphasis on technical issues, a low level of maturity for ad-hoc projects and the ambiguity of roles in data science are among these challenges. Few methodologies have been proposed on the literature that tackle these type of challenges, some of them date back to the mid-1990, and consequently they are not updated to the current paradigm and the latest developments in big data and machine learning technologies. In addition, fewer methodologies offer a complete guideline across team, project and data & information management. In this article we would like to explore the necessity of developing a more holistic approach for carrying out data science projects. We first review methodologies that have been presented on the literature to work on data science projects and classify them according to the their focus: project, team, data and information management. Finally, we propose a conceptual framework containing general characteristics that a methodology for managing data science projects with a holistic point of view should have. This framework can be used by other researchers as a roadmap for the design of new data science methodologies or the updating of existing ones.