The problem with data science education

THE demand for data science professionals has been echoed in the increased supply of data science education. In the field of data science alone, we have witnessed a flourishing of formal degrees, diplomas, and certifications, among other forms of training.

However, has our eagerness in producing data scientists resulted in the proliferation of generalists rather than practical experts in the field?

What is data science? Unfortunately, this is a complex question with a diverse answer depending on who you ask. Interestingly, it is the diversity of this answer that results in the quandary as to the skills required to be acquired by a data science professional.

According to Wikipedia, data science is, in the broad sense, "an inter-disciplinary field that uses scientific methods, processes, algorithms, and systems to extract knowledge and insights from many structural and unstructured data".

The field is a hybrid of statistics, machine learning, and data mining, and covers programming languages, technology frameworks, development platforms, as well as visualisation tools.

In other words, data science is the umbrella under which all these individual and/or intersecting disciplines sit.

The diversification and complexity of data science poses a real academic challenge in data science education. How does one balance the multiplicity of areas, covering them with sufficient depth in order to produce a well-rounded data scientist who can provide impactful value to a future employer?

Producing effective data scientists

The article Placing 'practice' at the centre of data science education by Eric Kolaczyk, Haviland Wright, and Masanao Yajima highlighted such a challenge where existing degree programmes "can leave students, upon exiting academia, needing a non-trivial ramping-up period before they can truly have an impact with their first employers".

This feedback is common and goes beyond that of the applicability of the practicum component of a degree. The majority of data scientists, upon completion of their education, still require a hands-on education at their place of employment to refine their learned skills to practical necessities.

In other words, a freshly minted, well-rounded data scientist is unlikely to hit the ground running at their first employ despite a presumption otherwise.

How then should well-rounded data scientists be produced - so that they can be effective, as close as possible to day one, upon exiting their respective training?

Rather than trying to cover the breadth of all that is data science, academia may wish to recognise that the field has matured and evolved in its pedagogy taxonomy.

Focusing on professional application, rather than pure academic research, we can identify the following four key example roles that an organisation may need in operationalising their data science ambitions:

• The data science developer (or modeller), who would be the core developer with a stronger focus on data science methodologies and techniques.
• The data science engineer, who would focus on data piping, data quality, and data ingestion for the purpose of data science modelling.
• The data science solution architect, who would have a deeper understanding of platforms and data enterprise architecture for an end-to-end operationalisation.
• The data science storyteller, who would have a stronger focus on data visualisation and data science communication, and possess strong business process acumen.

The aforementioned roles are not just an application of data science. While having overlapping areas, each carries an in-depth focus, specific required competencies, skill sets, and practicum considerations.

These elements highlight that a unified data science programme may not do any of the respective areas justice, as analytics author Thomas Davenport suggested. Worse, an employer may hire using the generic description of data scientist without realising they may need one specific area over the others - or, indeed, require four separate hires rather than the one.

Refocusing efforts

Whether through the creation of dedicated degrees with specific focus areas, diversification of modules that allow the flexibility of commonality and a major/minor approach, or a degree that differentiates between a research and a practicum (professional) focus, it is clear that there is a need to refocus our efforts.

Data science academic programmes should tailor data science education to produce professional competency with sufficient depth that enables immediate demonstrable outcome upon completion. Such programmes need to reflect the maturity of the data science field and allow more specificity in the professional expertise they aspire to produce.

Alternatively, or perhaps concurrently, employers need to appreciate that newly minted data scientists are not quite "road-ready" and may require secondary on-the-ground training in order to contextualise and surface the benefits anticipated from data scientists.

• The writer is senior adviser for data and artificial intelligence (AI) at UnionBank Philippines. He is concurrently an external adviser to Singapore's Corrupt Investigation Practices Bureau (on AI) and to the Central Provident Fund Board (on data science). He was previously the Monetary Authority of Singapore's first chief data officer.