Towards a Conceptual Framework for the Specification of Reproducible and Replicable Data Analysis Projects

Abstract

It is becoming increasingly common to exploit the data collected by Information Systems in order to carry out an analysis of them and obtain conclusions that give rise to a series of decisions in the different research fields. The fact that in most cases these conclusions cannot be properly backed up has given rise to a reproducibility crisis in Data Science, the discipline that makes it possible to convert such data into knowledge, and it research fields that apply it. In this paper we envision a conceptual framework to foster reproducible and replicable Data Science projects. The framework proposes the definition of systematic pipelines that may be (semi)automatically executed in terms of concrete implementation platforms. Proprietary or third party tools are also considered so that flexibility may be ensured without hindering reproducibility and replicability.