A practical guide for data science practices leveraging established data science models

Abstract

In the existing literature, a plethora of models offer guidance for data mining projects. Notably, CRoss-Industry Standard Process for Data Mining (CRISP-DM), established itself as the de facto standard for the development of projects in data mining and knowledge discovery. This status persists and has been proved by various surveys and polls conducted over the past two decades since its introduction.

Literature identifies several shortcomings in CRISP-DM and other models, including inadequate project management support and overly complex task structures. Further, with the dominance of data science over data mining, driven by technological advancements and evolving analytics practices, numerous research encourages modifications to the CRISP-DM model to better align with current needs. In the literature, new models such as Snail Shell, IKDDM, and DST have been introduced to address various limitations of CRISP-DM. However, none of these models successfully encompass the three key components: process, methodology, and lifecycle of data science projects.

This thesis introduces a new guideline, named ‘Data Science Executable Guide (DSEG)’ as a solution to address the identified gaps in existing literature, utilizing a Design Science Research (DSR) methodology. In addition to the limitations identified through literature review, requirements were gathered through interviews with a case company to develop the DSEG. The DSEG was applied in the case company scenario to validate its effectiveness. DSEG aims to address shortcomings in existing models by focusing on project management, flexibility in data science projects, and simplicity. It comprises seven steps and allows for limited iterations, thus mitigating complexity. Additionally, due to its characteristics, DSEG encompasses all key components (process, methodology, and lifecycle) of data science projects.