In 2009 Google published a short paper showing it could track the spread of flu across the United States faster than the health authorities — not by testing anyone, but by watching what people typed into a search box. Millions of queries for "sore throat" and "fever remedy," turned into a map. That is data science in one sentence: taking the exhaust of the modern world — clicks, transactions, sensor readings, medical records — and squeezing an answer out of it.
More carefully: data science is the discipline of turning data into decisions. It sits at the meeting point of three older subjects, and it needs all three at once.
The classic picture of data science is a Venn diagram. A data scientist lives in the middle, where mathematics & statistics, computing and domain knowledge all overlap.
Drop any one circle and the work goes wrong in a predictable way. Statistics without computing never touches real data; computing without statistics fools itself; either without domain knowledge answers the wrong question beautifully.
These three overlap so much that people argue about the borders. A rough, useful map:
| Field | Central question | Typical output |
|---|---|---|
| Statistics | Is this effect real, and how sure are we? | An estimate with error bars; a p-value; a confidence interval |
| Machine learning | Can we predict the next case accurately? | A trained model that generalises to new data |
| Data science | What should we do, given all this data? | A decision, a dashboard, a product feature, a report |
Data science is the widest of the three: it borrows the rigour of statistics and the predictive muscle of machine learning, then wraps them in the messy realities of finding data, cleaning it, and persuading someone to act. It cares about the whole journey from a raw CSV to a decision — which is exactly the journey this course follows, module by module.
In 2012 the Harvard Business Review ran an article titled "Data Scientist: The Sexiest Job of the 21st Century," and the label stuck for a decade. The phrase was half a joke, but the demand behind it was real: organisations had suddenly accumulated vastly more data than they knew how to use, and almost nobody could bridge statistics and software at once. The interesting twist is what happened next — the role fragmented. Today's "data scientist," "ML engineer," "data analyst," "data engineer" and "research scientist" were once one impossibly broad job description. Knowing the maths in this course is what lets you move between all of them.
The single most common failure in data science is treating a pattern in the data as an answer. The
number of films Nicolas Cage appears in correlates almost perfectly with the number of people who
drown in swimming pools each year — a real, strong correlation, and utterly meaningless. Data is
full of these coincidences, and the more variables you have, the more of them you will find by pure
chance. The whole apparatus of statistics — sampling, significance testing, cross-validation, causal
reasoning — exists to stop you shipping a spurious pattern as a fact. A data scientist who forgets
this is just an expensive random-pattern generator. We return to this problem in depth in the module
on
We follow the real shape of a data-science project. First we get and tidy the data (Modules 1–2), then explore it (Module 3) and engineer features from it (Module 4). Then we model and validate honestly (Modules 5–7), reason about experiments and cause (Module 8), handle time (Module 9), and finally make results interpretable, fair and persuasive (Module 10). Every step is taught as mathematics; the matching Python practice lives on Kaggle.
This course teaches the ideas; the code is best learned by doing. If you have never written Python, start with Kaggle Learn's free, in-browser Intro to Programming and Python courses, then move to Pandas to start handling real tables of data. You'll practise there what we explain here.