Data science sits behind recommendations, fraud alerts, demand forecasts and, more recently, the AI assistants many of us use daily. This guide explains where the field came from, what the work involves, and gives a complete learning roadmap with a stage-by-stage syllabus, a toolkit, portfolio projects and an honest look at where the field is heading. Facts and figures are linked to their sources at the end; where something is our own planning estimate or opinion, we say so.
What is data science?
Data science is an interdisciplinary field that combines statistics, scientific computing, algorithms and programming to extract knowledge and insight from data, whether that data is neat tables or messy text, images and logs. In practice a data scientist turns a business question into a data question, prepares the data, builds and tests a model or analysis, and then communicates a result that someone can act on.
The most useful way to think about the work is as a loop rather than a straight line:
- Frame the problem. What decision will this change, and how will we know it worked?
- Collect and clean data. Find it, join it, check its quality, handle gaps and errors.
- Explore. Summarise and visualise to understand patterns, outliers and risks.
- Model. Choose methods, train, validate honestly and compare against a simple baseline.
- Communicate and deploy. Explain results clearly, put the model or dashboard where people can use it, then monitor it.
A short history
The ideas are older than the job title. These are the milestones usually cited, with dates as recorded in the sources listed below.
| Year | Milestone |
|---|---|
| 1962 | John Tukey describes "data analysis", a discipline that resembles what we now call data science. |
| 1974 | Peter Naur uses "data science" as an alternative name for computer science. |
| 1985 | C. F. Jeff Wu uses "data science" in a lecture as an alternative name for statistics. |
| 2001 | William S. Cleveland publishes an action plan for expanding the technical areas of statistics, often linked to the modern idea of data science as its own discipline. |
| 2008 | The job title "data scientist" is attributed to DJ Patil and Jeff Hammerbacher. |
| 2010 | scikit-learn, the Python machine learning library that began as a 2007 Google Summer of Code project, has its first public release. |
| 2011 | Apache Hadoop 1.0 is released, bringing large-scale distributed data processing to many organisations. |
| 2012 | Harvard Business Review calls data scientist "the sexiest job of the 21st century". The same year, AlexNet wins the ImageNet challenge with a 15.3% top-5 error rate, a result that kicked off the deep learning wave. |
| 2017 | Google researchers publish "Attention Is All You Need", introducing the Transformer architecture that underlies most large language models. |
| Late 2022 | Chat-style generative AI reaches the general public, moving data and AI skills into everyday conversation. |
Roles in the data world
Job titles overlap, and companies use them differently, so judge a role by its responsibilities rather than its name. Broadly:
| Role | Typical focus | Core skills |
|---|---|---|
| Data analyst | Answering business questions with reports and dashboards | SQL, spreadsheets, BI tools such as Tableau or Power BI, clear communication |
| Data scientist | Statistical analysis, experiments and predictive models | Statistics, Python or R, machine learning, experimentation |
| Data engineer | Building and maintaining the pipelines and platforms data depends on | Strong programming, SQL, cloud platforms, data modelling |
| Analytics engineer | Bridging engineering and analysis with well-modelled, tested data | SQL, software engineering practice, warehouse tools |
| Machine learning engineer | Designing and deploying machine learning systems | Python, ML frameworks, software engineering, deployment |
A related discipline is MLOps, which Google Cloud describes as a practice that unifies machine learning system development with its operation, covering automated pipelines, data and model validation, and monitoring in production.
The complete roadmap: seven stages
The plan below suits a motivated beginner studying roughly ten hours a week. The durations are our own planning estimates, not guarantees; people with a strong maths or programming background will move faster, and that is fine. Some structured programmes publish similar pacing: for example, IBM's Data Science Professional Certificate on Coursera lists 12 courses over about four months at ten hours a week, and Google's Data Analytics Certificate lists nine courses over about six months at the same pace. Those are useful reference points for what a beginner-level programme looks like, and neither is a promise of a job.
| Stage | Focus | Estimated time |
|---|---|---|
| 0 | Foundations: maths refresher, computing basics, tools | 4 to 6 weeks |
| 1 | Python and SQL for data | 6 to 8 weeks |
| 2 | Statistics, exploration and visualisation | 6 to 8 weeks |
| 3 | Machine learning fundamentals | 8 to 10 weeks |
| 4 | Deep learning, NLP and time series | 8 to 10 weeks |
| 5 | Data engineering, deployment and MLOps | 6 to 8 weeks |
| 6 | Capstone project, portfolio and job preparation | 4 to 6 weeks |
That adds up to roughly ten to thirteen months at a steady pace. Treat it as a map, not a timetable.
Syllabus: topics to cover at each stage
Stage 0: Foundations
- Arithmetic, algebra and functions; reading graphs and equations without fear
- Introductory linear algebra: vectors, matrices and what a matrix multiplication does
- Introductory calculus: rates of change and why gradients matter for training models
- Probability basics: events, conditional probability, Bayes' rule
- Computing basics: files and folders, the command line, Git and GitHub, Jupyter notebooks
Stage 1: Python and SQL for data
- Python fundamentals: variables, control flow, functions, data structures, modules, error handling
- NumPy for arrays and vectorised computation
- pandas for loading, cleaning, reshaping, merging and aggregating data
- Reading data from CSV, Excel, JSON, APIs and basic web scraping (respecting site terms)
- SQL: SELECT, WHERE, GROUP BY, joins, subqueries, window functions, CTEs
- Data quality: missing values, duplicates, outliers, inconsistent formats
Stage 2: Statistics, exploration and visualisation
- Descriptive statistics, distributions, variance, correlation (and why correlation is not causation)
- Sampling, confidence intervals and hypothesis testing
- Regression: linear and logistic, interpreting coefficients
- Experiment design and A/B testing, including common pitfalls such as stopping early
- Exploratory data analysis as a repeatable routine
- Visualisation with Matplotlib, Seaborn or Plotly; choosing the right chart; dashboards in Tableau or Power BI
- Data storytelling: turning a finding into a one-page recommendation
Stage 3: Machine learning fundamentals
- The workflow: train, validation and test splits; baselines; cross-validation
- Supervised learning: linear and logistic models, decision trees, random forests, gradient boosting, k-nearest neighbours, support vector machines
- Unsupervised learning: clustering (k-means, hierarchical), dimensionality reduction (PCA)
- Feature engineering, encoding categorical variables, scaling, handling imbalanced classes
- Evaluation: accuracy, precision, recall, F1, ROC-AUC for classification; MAE and RMSE for regression; choosing the metric that matches the business cost
- Overfitting, regularisation and hyperparameter tuning
- Model interpretation and explaining predictions to non-technical stakeholders
- Tooling: scikit-learn pipelines
Stage 4: Deep learning, NLP and time series
- Neural network basics: layers, activation functions, loss, backpropagation, optimisers
- Frameworks: PyTorch or TensorFlow and Keras
- Computer vision: convolutional networks, transfer learning with pretrained models
- Natural language processing: text cleaning, embeddings, Transformers, fine-tuning and prompt-based workflows
- Working with large language models: retrieval-augmented generation, evaluation, cost and safety considerations
- Time series: trends, seasonality, forecasting baselines, classical models such as ARIMA, and machine learning approaches
- Recommendation systems at an introductory level
Stage 5: Data engineering, deployment and MLOps
- Data warehousing and modelling concepts; batch pipelines and scheduling
- Big data tools such as Apache Spark, and when you do not need them
- Cloud fundamentals on at least one major provider: storage, compute, managed databases
- Packaging a model behind an API; containers with Docker
- Version control for code, data and models; experiment tracking
- Continuous integration and delivery for machine learning; monitoring for data and model drift
- Reproducibility: environments, configuration and documentation
Stage 6: Capstone, portfolio and job preparation
- One end-to-end capstone: a real question, real data, a defensible method, a deployed demo and a written report
- A clean GitHub profile with documented repositories and readable READMEs
- Interview preparation: SQL and Python exercises, statistics questions, ML concepts, case-style problems
- Communication practice: presenting to a non-technical audience in five minutes
Cross-cutting topics you should not skip
- Ethics and fairness. Understand bias in data and models, privacy, consent and the real-world impact of automated decisions.
- Privacy and regulation. Data protection laws such as India's Digital Personal Data Protection Act, 2023 and the EU AI Act are changing how data may be collected and how AI systems must be governed. Their obligations are being phased in, so check the official texts for current dates and requirements.
- Domain knowledge. A model for healthcare, retail or finance needs an understanding of how that business works. Pick a domain you care about.
- Communication. Findings that cannot be explained do not get used.
The standard toolkit
- Languages: Python (primary), SQL (essential), R (common in statistics and research)
- Libraries: NumPy, pandas, scikit-learn, Matplotlib, Seaborn, Plotly, PyTorch or TensorFlow
- Environments: Jupyter, VS Code, Git and GitHub
- Data and BI: PostgreSQL or another relational database, Tableau or Power BI, Excel
- Scale and delivery: Spark, Docker, a cloud platform, workflow schedulers
Portfolio project ideas
Hiring teams value a small number of well-explained, end-to-end projects over a long list of tutorial copies. Overused datasets such as Titanic, Iris and MNIST do not set you apart. Better ideas:
- Business analytics: analyse a public retail or transport dataset in SQL, then build a dashboard that answers three concrete questions.
- Prediction: predict customer churn or late deliveries; compare against a simple baseline and justify your metric.
- Forecasting: forecast demand or energy use with proper backtesting.
- NLP: classify or summarise support tickets or reviews, and document where the model fails.
- Deployment: wrap one of the above in an API with a simple web front end and a monitoring note.
- Local problem: use open data about your city or college and write up what you found and what you could not conclude.
For each project, explain the question, the data and its limits, your method, how you validated it and what decision it would support.
Common beginner mistakes
- Jumping to complex models before mastering SQL, pandas and basic statistics
- Leaking information from the test set into training, which makes results look better than they are
- Tuning to a metric that does not match the real cost of an error
- Collecting certificates instead of building and explaining projects
- Skipping the problem-framing step and starting with the algorithm
- Never deploying anything, so the work never meets real data or users
Careers and outlook
Demand figures vary by country and source, so here are the ones we can point to directly.
- United States: the Bureau of Labor Statistics reports a median pay of $120,230 per year for data scientists (May 2025 data) and projects employment to grow 35 percent from 2025 to 2035, which it describes as much faster than the average for all occupations, with about 24,800 openings a year on average. These are US figures and should not be read as Indian salaries or hiring numbers.
- Global: the World Economic Forum's Future of Jobs Report 2025, as reported in coverage of it, lists big data specialists and AI and machine learning specialists among the fastest-growing roles in percentage terms for 2025 to 2030.
- India: we do not quote a headline figure here because we could not verify a current, official one. Check recent NASSCOM publications for up-to-date numbers.
Pay and demand depend heavily on location, experience and the strength of your portfolio. No course, including ours, can promise a job or a salary.
Where the field is heading: our view
The points below are our own assessment, not statistics.
- Generative AI is becoming part of the toolkit. Working with large language models, retrieval and evaluation is increasingly part of a data professional's job, and the line between data science and AI engineering is blurring.
- Automation takes over routine steps. Automated machine learning and coding assistants speed up routine modelling and boilerplate, which raises the value of problem framing, judgement and verification.
- Data quality gets more attention. Better, cleaner and better-documented data often beats a fancier algorithm.
- Production skills matter more. Employers increasingly want people who can deploy, monitor and maintain models, not only build prototypes.
- Responsible AI becomes a requirement. Regulation, privacy law and customer expectations push teams to document, test and explain their systems.
- The fundamentals do not expire. Statistics, SQL, clear thinking and communication stay valuable whichever tools are fashionable this year.
How to start this week
- Install Python and Jupyter, and create a GitHub account.
- Pick one small dataset about something you care about and ask three questions of it.
- Learn enough SQL to answer those questions with a query.
- Write up what you found in a short notebook. That is your first portfolio piece.
- Choose a structured path with feedback, mentors and projects, so you are not studying alone.
If you would like guided, project-based learning, explore the courses at Nexelence Academy or talk to us about the right starting point for your background.
Sources
- History of the term and definition: Wikipedia, Data science (check the primary papers cited there for the original sources)
- Davenport and Patil, Data Scientist: The Sexiest Job of the 21st Century, Harvard Business Review, October 2012
- Apache Hadoop, scikit-learn, AlexNet, Attention Is All You Need (milestone dates and results)
- Google Cloud, MLOps: continuous delivery and automation pipelines in machine learning
- IBM, Data Science Professional Certificate; Google, Data Analytics Professional Certificate (published pacing and course counts)
- US Bureau of Labor Statistics, Occupational Outlook Handbook: Data Scientists (page updated August 2026)
- World Economic Forum, Future of Jobs Report 2025 (figures here are from secondary coverage of the report)
- EU AI Act implementation timeline; India's Digital Personal Data Protection Act, 2023 and its rules (confirm current dates in the official texts)
- #data science
- #roadmap
- #syllabus
- #machine learning
- #python
- #sql
- #career