Skip to content
Tutorials Center
My learning

Topics

Lesson 1 of 76 ยท What Data Science Is

What Is Data Science?

Learn what data science is, what data scientists actually do, how a project goes from question to decision, and run a tiny end-to-end analysis in Python.

Data science is the practice of using data to answer questions and support decisions. A shop wants to know which customers are about to leave. A hospital wants to know which patients need a follow-up call. A website wants to know whether a new button color really increases sign-ups. In each case someone has a question, some data that might help, and a decision to make. Data science is the craft that connects those three things.

It is not one skill. It mixes programming, statistics, domain knowledge and communication. A good analysis can be a ten-line script and a clear sentence, while a bad one can be a huge model that answers the wrong question. This course teaches you the whole path in a practical order: asking good questions, getting and cleaning data, exploring it, drawing honest conclusions, building models and sharing results.

The three kinds of questions

Almost every data question falls into one of three families. Knowing which one you are facing tells you which tools to reach for.

Kind Question Example Typical tools
Descriptive What happened? Which product sold best last month? Grouping, summaries, charts
Predictive What will happen? Which customers will cancel next month? Models, validation
Causal What would happen if we did X? Does free shipping increase orders? Experiments, careful design

The families build on each other, but they are not interchangeable. A model can predict that customers who got a coupon buy more, yet that does not prove the coupon caused the purchases: maybe the marketing team simply sent coupons to the keenest customers. Keeping the three kinds apart is one of the most valuable habits you can build, and Causal Thinking and Confounding comes back to it in detail.

From question to decision

Real projects follow a loop rather than a straight line. You start with a question, collect data, clean it, look at it, build something, check it, and then use the result. Often the check sends you back to an earlier step.

flowchart LR
    Q["Question"] --> D["Get data"]
    D --> C["Clean and prepare"]
    C --> E["Explore"]
    E --> M["Model or test"]
    M --> V["Check and explain"]
    V --> A["Decision or action"]
    V -.->|"not good enough"| Q

The data science loop: most of the real work happens in the middle steps, and weak results send you back to the start.

The next lesson, The Data Science Workflow, looks at each box in more detail. For now, notice that "build a model" is only one box out of seven.

Who does what

Job titles vary between companies, but the work falls into a few recognizable roles:

  • Data analyst answers descriptive questions with queries, dashboards and short reports.
  • Data scientist adds statistics and modelling to answer predictive and causal questions.
  • Machine learning engineer turns models into reliable software that runs in production.
  • Data engineer builds the pipelines and databases that deliver clean data to everyone else.
  • Domain expert or product manager knows what a good answer would actually change.

On a small team one person wears several hats. That is one reason this course covers a broad range, from loading a file to deploying a model.

The toolbox

This tutorial uses Python, because its libraries cover the whole workflow and every example runs live in your browser. You will meet these tools again and again:

  • NumPy stores and computes with arrays of numbers. See the NumPy tutorial.
  • pandas holds tables of data and makes cleaning, grouping and joining easy. See the pandas tutorial.
  • SciPy and statsmodels provide statistical tests and models. See What Is SciPy?.
  • matplotlib draws the charts.
  • scikit-learn offers machine learning models and tools to evaluate them.

You do not need to master them all first. If you can write simple Python, you can follow along and learn the libraries as they appear. The Python tutorial has a short lesson on using pandas if you want a first taste.

A tiny end-to-end analysis

Here is a complete miniature analysis in about twenty lines. The question: do customers who receive a reminder email spend more? We make up a small dataset with a fixed random seed, so the numbers are the same every time you run it.

Python A first end-to-end analysis

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

In a few lines you did what a real project does at small scale: created or loaded data, summarized it by group and compared the groups. What is missing is the honest part. Is a difference of that size just luck? Could another factor explain it? Statistical testing, confidence intervals and experiment design answer those questions in later chapters.

Adding a picture and a sentence

A result nobody understands changes nothing. The last step of almost every project is a chart plus one clear sentence. Here is the same analysis with both:

Python Chart and a one-sentence finding

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Notice the careful wording: "in this sample". A good finding says what was measured and does not claim more than the data supports.

Be aware that the spending data here is simulated. In this course we often make up data with a known recipe so you can see whether a method recovers the truth. Real data is messier, which is why later chapters spend so much time on cleaning and checking.

Why data science matters and what it is built from

Organizations collect far more data than people can read, and the cost of storing and processing it keeps falling, so the demand for people who can turn it into decisions keeps growing. The benefits show up in a few repeating forms:

  • Better decisions and forecasting: replace gut feeling with measured evidence, and plan stock, staffing and budgets from forecasts.
  • Efficiency and optimization: find the bottleneck in a process, schedule routes, cut waste, detect equipment trouble before it stops production.
  • Personalization: recommend what each customer is likely to want, and tailor messages, prices or learning paths.
  • Healthcare and public services: spot patients at risk earlier, allocate scarce resources, and evaluate whether a program works.
  • New products: search, recommendations, fraud screening and voice assistants are data products.

Under the job title sit a few building blocks that this course teaches in order:

Building block What it gives you Where in this course
Statistics and mathematics Describing uncertainty, testing claims, understanding models Descriptive Statistics, Probability and Distributions
Programming and tools Repeatable analysis in Python, SQL and notebooks The Data Science Workflow
Data wrangling Collecting, cleaning and reshaping messy data Data Quality Audit, Missing Values
Machine learning Learning patterns that predict Modelling Overview
Visualization and communication Showing results so people act on them Visualization Principles, Communicating Results
Domain knowledge Knowing which questions matter and which numbers make sense Framing the Question

The data science lifecycle strings these together: understand the problem, collect data, clean and explore it, build and evaluate a model (the data modelling step), deploy the result and monitor it. Each stage can send you back to an earlier one, which The Data Science Workflow describes in detail. A library map shows which Python tools serve which stage.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. A manager asks: 'Which of our customers are most likely to cancel next month?' What kind of question is this?

    Estimating what will happen in the future for each customer is a predictive question. It needs a model validated on past data.
  2. 2. A model shows that customers who received coupons spent more. What can you safely conclude?

    An association can come from confounding, for example if coupons were sent to the most active customers. A randomized experiment is the clean way to test cause.
  3. 3. Which role is mainly responsible for building the pipelines that deliver clean data to analysts?

    Data engineers build and maintain data pipelines and storage. Analysts and scientists consume what those pipelines produce.
  4. 4. Why does the data science loop have an arrow back to the question?

    Checking results can reveal that the question was vague, the data was wrong or the approach does not fit, so you iterate.

Summary

  • Data science uses data to answer questions and support decisions; it blends programming, statistics, domain knowledge and communication.
  • Questions are descriptive (what happened), predictive (what will happen) or causal (what if we act).
  • Projects are loops: question, data, cleaning, exploration, modelling or testing, checking, decision.
  • The main Python tools are NumPy, pandas, SciPy and statsmodels, matplotlib and scikit-learn.
  • A finished result pairs a chart with an honest one-sentence finding.

Next: The Data Science Workflow, a closer look at each stage and where the time really goes.