Data science skills in 2026: what job posts really ask for

Data science skills are the skills you need to turn raw data into answers people can use. The core set is Python, SQL, statistics, machine learning, some cloud basics and clear communication.

We use real job-post data, not a course advert. You get the numbers, what the numbers cannot tell you, a learning order for students and ways to prove your skills with projects. All facts are as of November 2026.

What are the core data science skills?

The core data science skills fall into six groups. Most data scientist job posts ask for some mix of these:

• Python: the main language for data work. You use it to clean data, run analysis and build models, with libraries such as pandas and scikit-learn.
• SQL: the language for getting data out of databases. Most company data lives in tables, and SQL is how you reach it.
• Statistics: probability, sampling, hypothesis tests and regression. Statistics tells you if a result is real or just noise.
• Machine learning: models that learn patterns from data to predict or group things, such as regression, decision trees and clustering.
• Cloud basics: running your work on platforms such as AWS or Microsoft Azure, where company data and computers often live.
• Communication: explaining what you found, and how sure you are, to people who are not data scientists.

What does job-post data say about data science skills?

Job-post data puts machine learning and Python at the top. The training company 365 Data Science published a study in April 2026. It is based on 1,000 Monster job posts (827 analysed) for data scientists. Here is how often each skill appeared:

• Machine learning: 69% of posts.
• Python: 57%, down from 78% in their earlier study.
• R: 33%.
• SQL: 30%, down from 60% in their earlier study.
• AWS (cloud): 19.7%.
• NLP: 19%, up from 5% in their earlier study. NLP (natural language processing) means working with text.
• ML libraries: PyTorch 10.8%, TensorFlow 10.4%, scikit-learn 6.4%.
• Dashboards: Tableau 11.5%, Power BI 9.1%.
• Excel: 10%, down from 26% in their earlier study.

The same study looked at degrees. 20% of posts asked for a bachelor's degree, 30% for a master's and 24% for a PhD. Another 26% did not say.

One number stands out. "Statistical analysis" was named in only 7.7% of posts. Yet the same article says that "statistics knowledge and solid data analysis skills form the foundation of data science expertise." Employers often expect statistics without writing it down.

The article also says where hiring is going. In its words, "the strongest candidates will be those who can evaluate models critically, explain results clearly, and connect technical work to business impact." That is communication and statistics, not a new tool.

Is Python or SQL more important for data science?

You need both. Python is asked for more often in data scientist job posts, but among developers SQL is used just as widely.

In the Stack Overflow Developer Survey 2026, 14,099 people answered which languages they had done extensive work in over the past year. SQL was used by 58.4% of them and Python by 58.0%. R was used by 4.5%.

This survey covers all kinds of developers, not just data scientists. Still, the message is clear: SQL is everywhere. A job post may skip SQL simply because the employer takes it for granted.

Our advice: learn Python first, then SQL right after. Learn R later if your target jobs or your research group use it.

How is AI changing data science jobs in India?

AI skills are now common in Indian tech job descriptions. The Indian job site Naukri published a report in July 2026. It found that AI mentions in technology job descriptions grew from 0.9% in 2020 to 15% in 2026, nearly 17 times more.

The same Naukri report found that 77% of professionals use AI at work at least sometimes. But only 35% say their employer actively provides AI training. For students, this means you will most likely learn AI skills on your own, before your first job.

This matches the 365 study, where NLP rose from 5% to 19% of posts. So learn the basics of machine learning well, then learn how language models work. AI tools do not replace the core skills. You still need statistics to check if a model's answer is right.

What math do you need for data science?

You need statistics and probability the most, plus some linear algebra and basic calculus for machine learning.

• Statistics: averages, spread, sampling, confidence intervals, hypothesis tests and regression.
• Probability: chance, distributions and conditional probability.
• Linear algebra: vectors and matrices. Data is stored as tables of numbers, and models work on those tables.
• Calculus: derivatives, enough to see how a model learns by reducing its error step by step.

Many students rush to machine learning and skip statistics. That is a mistake, because statistics is what stops you from fooling yourself with a lucky result. Our Statistics for Researchers course covers p-values, confidence intervals, effect size and regression in simple words.

Do you need a degree to become a data scientist?

Most data scientist job posts ask for a degree, but not all. In the 365 Data Science study, 74% of posts asked for a bachelor's, master's or PhD, and 26% did not state a degree.

A degree in any subject with numbers, such as engineering, maths, statistics or computer science, helps. If you do not have the "right" degree, finished projects are how you show what you can do.

How much can you trust these numbers?

Treat skill rankings as a rough guide, not exact truth. Every source in this post has limits:

• 365 study: the page does not say which country the posts came from or when they were posted. It also labels its years in different ways, and 365 Data Science sells data science courses.
• Job posts: a post lists what an employer wrote, not what the job uses every day. Results from 827 posts can change a lot from one study to the next.
• Stack Overflow: the survey covers all kinds of developers, and people choose to take it.
• Naukri: the 15% figure counts AI mentions in all tech job descriptions, not just data science jobs.

So do not chase one percentage. Look at what stays near the top across sources: Python, SQL and machine learning. Statistics sits under all three, even when posts do not name it.

What is the best data science roadmap for beginners?

A good data science roadmap for beginners goes from tools, to methods, to proof. Here is the order we suggest for a student. Finish one small project at each step before you move on.

1. 1Python and pandas: learn variables, loops and functions, then load a CSV file and clean it. One free option is the NPTEL course Python for Data Science from IIT Madras. NPTEL says course learning is free and the certificate exam is optional.
2. 2SQL: learn SELECT, WHERE, GROUP BY and JOIN. Practise on a free database such as SQLite.
3. 3Statistics: learn to describe data, measure uncertainty and test an idea. Then re-analyse a small open dataset.
4. 4Cleaning and charts: handle missing values, fix wrong types and plot your data with matplotlib or seaborn. Real data is messy, so this step needs practice.
5. 5Machine learning: start with scikit-learn. Learn linear and logistic regression, decision trees, and how to split data into training and test sets.
6. 6Git and GitHub: save every project in a public repository with a README that explains how to run it.
7. 7Cloud basics: run one notebook on one cloud platform. Learn one platform well, not three a little.
8. 8Write it up: after each project, write one page with the question, the data, the method, the result and its limits.
9. 9Specialise: pick one area, such as NLP, deep learning with PyTorch, or time series, and go deeper.

How do you prove data science skills with projects?

You prove data science skills with finished projects that others can check: public code, clear data and a short write-up. A certificate shows you studied. A project shows you can do the work.

A strong student project has five parts:

• A real question: "Which factors predict late deliveries in this dataset?" is better than "I trained a model".
• Public data: use open sources such as data.gov.in, India's Open Government Data Platform, so others can get the same data.
• Clean code: on GitHub, with a README that says how to run it.
• Honest results: compare with a simple baseline, and say where the model fails.
• A write-up: one or two pages in plain language, or a short blog post.

Why do research projects count as proof?

Research projects are strong proof because they use every core skill at once. You ask a question, find data, pick a method, test the result with statistics and explain what you found. That is close to what the 365 study says employers want: people who can "evaluate models critically" and "explain results clearly".

An easy way to start is to reproduce a published paper. Take its code and data, run it, and check if you get the same numbers. Then change one thing and see what happens. Our Reproducible Research course shows how to set up a project so anyone can rerun it. For ideas, see our list of AI and ML research topics for students.

At APJ-RAISE, students do small research projects like these in teams and share the code and write-up on our Open Research page. Joining is free for students aged 18 or older.