The 12-month data scientist roadmap: Months 1–3 — Python, pandas, statistics, and probability. Months 4–6 — SQL, scikit-learn, supervised/unsupervised learning, and feature engineering. Months 7–9 — deep learning with PyTorch or TensorFlow, plus one specialization (NLP, computer vision, or time series). Months 10–12 — build 3 portfolio projects, deploy a model, polish GitHub, and execute a targeted job search. Each phase produces a tangible project. The order matters — statistics before ML, ML before deep learning, not the other way around.
This article was researched and written by the Careery team — that helps land higher-paying jobs faster than ever! Learn more about Careery →
How long does it take to become a data scientist from scratch?
With a structured plan: 12 months studying 15–20 hours per week (part-time), or 6–9 months at 35–40 hours per week (full-time). The critical path is Python + statistics (3 months) → ML + SQL (3 months) → deep learning + specialization (3 months) → portfolio + job search (3 months). Career changers with a strong math or CS background can compress the timeline to 6–9 months by accelerating the statistics phase.
What should I learn first to become a data scientist?
Python and statistics — not machine learning, not deep learning, not Spark. Python (specifically pandas and NumPy) is the daily working language for 87%+ of data scientists according to the Kaggle Survey. Statistics provides the theoretical foundation that separates data scientists from people who just call scikit-learn functions. Build fluency in both before touching any ML library.
Can I become a data scientist without a master's degree?
Yes. A portfolio with 3 end-to-end projects — including at least one deployed model — demonstrates more applied skill than a diploma alone. That said, a master's degree still appears in roughly 60% of data science job postings as 'preferred.' The workaround: portfolio projects that show you can frame a business problem, build a model, evaluate it rigorously, and communicate findings. That's what the degree is supposed to prove.
Most aspiring data scientists don't fail because the math is too hard. They fail because they study the wrong things in the wrong order. Jumping into TensorFlow before understanding linear regression. Taking a deep learning course before knowing how to clean a dataset. Building a portfolio with Titanic and Iris — the same projects every other beginner submits.
This roadmap fixes that. Twelve months, four phases, each producing a project that belongs in a portfolio. By Month 12, the skills are proven, the GitHub is polished, and the job search is strategic — not desperate.
Data science is not data analytics with fancier tools. It requires mathematical thinking — the ability to reason about probability, optimization, and uncertainty. That doesn't mean a math degree is required. It means a realistic self-assessment before Month 1 saves months of frustration later.
Math background check: Comfort with algebra and basic calculus is the minimum. If "take the derivative of x²" feels foreign, spend 2–4 weeks on Khan Academy's precalculus and intro calculus courses before starting Month 1. Probability and statistics are taught in the roadmap, but they build on algebraic fluency.
Programming exposure: Zero programming experience is fine — Python is designed for readability, and Month 1 starts from scratch. Prior exposure to any language (JavaScript, R, even VBA) accelerates Months 1–2 significantly.
Time commitment: This roadmap assumes 15–20 hours per week. That's 2–3 hours on weekday evenings and 5–6 hours on weekends. Full-time learners (35–40 hours/week) can compress the timeline to 6–9 months. Career changers with strong math or CS backgrounds can skip ahead through the statistics phase and finish in 6–9 months at part-time pace.
This roadmap covers the learning path. For the full picture — including education options, career progression, day-to-day responsibilities, and salary expectations — see How to Become a Data Scientist in 2026.
Data science requires mathematical fluency that data analytics does not. Before starting Month 1, verify comfort with basic algebra and calculus. Zero programming experience is fine — but skipping the math foundation creates gaps that surface during ML and interviews.
The prerequisite check is done. Here's the month-by-month plan.
| Phase | Months | Focus | Key Deliverable | Hours/Week |
|---|---|---|---|---|
| Phase 1 | 1–3 | Python + Statistics Foundation | EDA notebook on a real-world dataset with statistical analysis | 15–20 |
| Phase 2 | 4–6 | Machine Learning + SQL | End-to-end classification or regression project with cross-validation | 15–20 |
| Phase 3 | 7–9 | Deep Learning + Specialization | Specialization project (NLP, CV, or time series) with trained neural network | 15–20 |
| Phase 4 | 10–12 | Portfolio + Job Search | 3 polished projects, deployed model, optimized resume, active applications | 20–25 |
Twelve months, four phases, four deliverables. Each phase produces a portfolio piece. The order is non-negotiable: Python and statistics first because they're the foundation, ML second because it depends on both, deep learning third because it depends on ML, portfolio and job search last because they depend on everything.
Here's what each phase looks like in detail.
Every data scientist's daily toolkit is Python. Not R, not Julia, not MATLAB — Python. The Kaggle Survey consistently shows 87%+ of data scientists use Python as their primary language. Months 1–3 build fluency in Python for data work and the statistical thinking that separates data science from software engineering.
Month 1: Python + pandas Fundamentals
Skills to learn:
- Python basics: variables, loops, conditionals, functions, list comprehensions
- pandas: reading CSVs, DataFrames, filtering, groupby, merge, pivot tables
- NumPy: arrays, vectorized operations, basic linear algebra
- Jupyter Notebooks for combining code, output, and documentation
Tools: Python 3.11+, Jupyter Lab, pandas, NumPy, Anaconda or miniconda
Resources:
- Python for Data Analysis by Wes McKinney (O'Reilly, 3rd edition, 2022) — written by the creator of pandas
- Kaggle's free Python and pandas micro-courses (browser-based, no setup)
Project milestone: A Jupyter Notebook that loads a real dataset (not Titanic or Iris — use something from data.gov or Kaggle Datasets), cleans it, and performs basic exploration with pandas. Publish on GitHub with a README.
You know you're ready when: Given a new CSV, you can load it, handle missing values, filter rows, group by categories, and produce summary statistics in under 30 minutes — without consulting documentation for basic operations.
Month 2: Probability + Descriptive Statistics
Skills to learn:
- Descriptive statistics: mean, median, mode, standard deviation, percentiles, IQR
- Probability fundamentals: conditional probability, Bayes' theorem, distributions (normal, binomial, Poisson)
- Data visualization: matplotlib basics, seaborn for statistical plots (histograms, box plots, pair plots, heatmaps)
- Exploratory data analysis (EDA) workflows
Tools: matplotlib, seaborn, scipy.stats
Resources:
- Khan Academy Statistics & Probability (free, structured curriculum)
- Think Stats by Allen Downey (free online) — Python-first statistics
You know you're ready when: You can explain the difference between population and sample statistics, calculate a confidence interval by hand, and produce a 10-chart EDA of any dataset that tells a coherent story.
Month 3: Inferential Statistics + Hypothesis Testing
Skills to learn:
- Hypothesis testing: null/alternative hypotheses, p-values, Type I/II errors
- t-tests, chi-squared tests, ANOVA — when to use each
- Correlation vs. causation (the most important statistical concept for data scientists)
- A/B testing fundamentals: experiment design, statistical significance, sample size calculations
- Linear regression as a statistical model (not just an ML algorithm)
Tools: scipy.stats, statsmodels
Phase 1 deliverable: A complete EDA notebook on a real-world dataset (at least 10,000 rows) that includes data cleaning, descriptive statistics, visualizations, hypothesis tests, and written interpretations of findings. This is the first portfolio piece — it should read like a data story, not a homework assignment. Publish on GitHub.
You know you're ready when: Someone shows you a dataset and a business question, and you can design the right statistical test, execute it in Python, and explain the results — including confidence intervals and effect sizes — to a non-technical audience.
- Spending 6 weeks on Python syntax before touching data — data scientists learn Python BY doing data work, not before it
- Skipping statistics to jump into scikit-learn — ML without statistics is just calling functions you don't understand
- Using only toy datasets (Iris, Titanic, MNIST) — employers want to see you work with messy, real-world data
- Watching 200 hours of tutorials without building anything — the skills don't stick without projects
Python and statistics are the non-negotiable foundation. Python is the language — 87%+ of data scientists use it daily. Statistics is the thinking — it determines whether model results are meaningful or noise. Spend 70% of Phase 1 on hands-on coding with real data, not watching lectures.
Phase 1 builds the foundation. Phase 2 turns that foundation into predictive power.
Machine learning is where data science differentiates itself from data analytics. But ML without the statistics foundation from Phase 1 is just pattern-matching with libraries — and interviewers can tell the difference. Phase 2 adds the tools that make data science data science.
Month 4: SQL + Data Access
Skills to learn:
- SQL fundamentals: SELECT, WHERE, GROUP BY, JOINs (INNER, LEFT, RIGHT)
- Advanced SQL: window functions (ROW_NUMBER, RANK, LAG, LEAD), CTEs, subqueries
- Database concepts: relational schemas, indexes, query optimization basics
- Connecting Python to databases: SQLAlchemy, pandas
read_sql()
Tools: PostgreSQL (or SQLite for local practice), DBeaver, SQLBolt
Resources:
- Mode Analytics SQL tutorial (uses real datasets)
- StrataScratch — real interview SQL questions from companies
You know you're ready when: You can write a multi-table JOIN with window functions and CTEs to answer a business question, and you can pull that same data into a pandas DataFrame for further analysis.
Month 5: Supervised Learning
Skills to learn:
- The ML workflow: train/test split, model fitting, prediction, evaluation
- Classification: logistic regression, decision trees, random forests, gradient boosting (XGBoost)
- Regression: linear regression, regularization (Lasso, Ridge), tree-based regressors
- Evaluation metrics: accuracy, precision, recall, F1, AUC-ROC, RMSE, MAE
- Cross-validation: k-fold, stratified k-fold, the bias-variance tradeoff
Tools: scikit-learn, XGBoost
Resources:
- Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow by Aurélien Géron (O'Reilly, 3rd edition, 2022) — the definitive ML reference
- Andrew Ng's Machine Learning Specialization on Coursera (updated 2022 version)
You know you're ready when: You can take a labeled dataset, preprocess it, train 3+ models, evaluate them with appropriate metrics, select the best one using cross-validation, and explain WHY it performs best — not just that it has the highest accuracy.
Month 6: Feature Engineering + Unsupervised Learning
Skills to learn:
- Feature engineering: encoding categoricals, scaling numerics, creating interaction features, handling datetime features
- Feature selection: correlation analysis, mutual information, recursive feature elimination
- Unsupervised learning: K-means clustering, hierarchical clustering, PCA for dimensionality reduction
- Pipeline building: scikit-learn Pipelines for reproducible workflows
Tools: scikit-learn, pandas, feature-engine
Phase 2 deliverable: An end-to-end classification or regression project on a real-world dataset (not from a course). Must include: problem framing, data cleaning, feature engineering, model selection with cross-validation, hyperparameter tuning, evaluation with appropriate metrics, and a written explanation of business insights. Publish on GitHub with a clear README.
You know you're ready when: Given a new dataset and a prediction task, you can build a complete ML pipeline — from raw data to tuned model — in under 4 hours, and explain every decision you made along the way.
For the full technical breakdown of every tool in the data science stack and what proficiency looks like at each level, see Data Science Skills and Tools.
Machine learning without SQL is impractical — real data lives in databases, not CSVs. SQL comes first in Phase 2 because every ML project starts with data access. The Phase 2 deliverable must demonstrate the full pipeline: data access, cleaning, feature engineering, model selection, evaluation, and business interpretation. Calling model.fit() is not data science — understanding why the model works is.
Phase 2 covers 80% of what entry-level data scientists do daily. Phase 3 adds the 20% that makes you competitive.
Deep learning appears in roughly 40% of data scientist job postings, and that number is climbing. More importantly, picking a specialization in Phase 3 — NLP, computer vision, or time series — gives the portfolio a focus that generic "I know a little of everything" candidates lack.
Month 7: Neural Network Fundamentals
Skills to learn:
- Neural network architecture: layers, activations, loss functions, backpropagation
- PyTorch OR TensorFlow basics (pick one — PyTorch has momentum in research and industry)
- Training workflows: batching, epochs, learning rate scheduling, early stopping
- Regularization: dropout, batch normalization, data augmentation
- GPU basics: using Google Colab or Kaggle notebooks for free GPU access
Tools: PyTorch (recommended) or TensorFlow/Keras, Google Colab
Resources:
- fast.ai Practical Deep Learning for Coders (free, top-down teaching approach)
- Hands-On Machine Learning by Géron, Part II (O'Reilly, 2022)
You know you're ready when: You can build, train, and evaluate a neural network from scratch in PyTorch/TensorFlow — and explain what each layer does, why the loss function was chosen, and how you'd debug underfitting or overfitting.
Months 8–9: Pick One Specialization
Choose one area to go deep. This is the differentiator — the thing that makes a portfolio memorable and a candidacy specific.
Option A — Natural Language Processing (NLP):
- Text preprocessing: tokenization, stemming, lemmatization, TF-IDF
- Word embeddings: Word2Vec, GloVe, contextual embeddings
- Transformer basics: attention mechanism, BERT, fine-tuning pretrained models with Hugging Face
- Project idea: sentiment analysis or text classification on a domain-specific dataset (not IMDB reviews)
Option B — Computer Vision (CV):
- Image preprocessing: resizing, normalization, augmentation
- CNNs: convolutional layers, pooling, architecture patterns (ResNet, EfficientNet)
- Transfer learning: fine-tuning pretrained models on custom datasets
- Project idea: image classification or object detection on a niche dataset (medical images, satellite data, manufacturing defects)
Option C — Time Series:
- Time series decomposition: trend, seasonality, residuals
- Classical methods: ARIMA, SARIMA, exponential smoothing
- ML for time series: feature engineering with lag variables, tree-based forecasting
- Deep learning for time series: LSTMs, Transformer-based forecasting
- Project idea: demand forecasting or anomaly detection on real business data
Phase 3 deliverable: A specialization project that demonstrates deep learning applied to the chosen domain. Must include a trained neural network, performance evaluation against a baseline model, and a clear narrative of why the approach was chosen. Bonus: deploy the model as an API (Flask or FastAPI) or interactive demo (Streamlit/Gradio).
You know you're ready when: You can explain the architecture choices in your specialization project, discuss the tradeoffs between approaches, and demonstrate that your model outperforms a simpler baseline — with metrics to prove it.
Specialization beats generalization for getting hired. A candidate who can say "I built an NLP pipeline that classifies customer support tickets with 94% accuracy" is more memorable than one who says "I know a little about NLP, CV, and time series." Pick one, go deep, and make it the centerpiece of the portfolio.
Three phases of skills. Phase 4 turns them into a job.
The difference between "studying data science" and "getting hired as a data scientist" is packaging and execution. Phase 4 converts nine months of skill-building into employment.
Month 10: Build and Polish 3 Portfolio Projects
By this point, there are already 3 project deliverables from Phases 1–3. Month 10 is about polishing them into hire-worthy portfolio pieces and filling any gaps.
The 3-project portfolio:
- EDA + Statistical Analysis (from Phase 1) — demonstrates data wrangling, visualization, and statistical reasoning on a real dataset
- End-to-End ML Project (from Phase 2) — demonstrates the full pipeline from raw data to tuned model with business interpretation
- Specialization Project (from Phase 3) — demonstrates deep learning expertise in a specific domain
Polish checklist for each project:
- Clear README: problem statement, approach, key findings, how to reproduce
- Clean code: well-commented, modular functions, requirements.txt
- Visualizations that tell a story, not just display data
- Business context: why this problem matters, what actions the results support
- GitHub repo with consistent formatting across all three projects
Bonus project: Deploy one model as a web app using Streamlit, Gradio, or FastAPI. A deployed model demonstrates production awareness — a skill most bootcamp graduates lack.
Month 11: Resume, LinkedIn, and Interview Prep
Actions:
- Build a tailored resume using the [problem → approach → tool → result] bullet formula
- Optimize LinkedIn: headline with specialization, summary with key projects, skills section with endorsements
- Practice ML interview questions: bias-variance tradeoff, regularization, evaluation metrics, A/B testing design
- Practice coding interviews: LeetCode Easy/Medium in Python (data structures, not algorithms-heavy)
- Prepare 3 project walkthroughs: 2-minute narratives covering problem, approach, result, and what you'd improve
Month 12: Job Search Execution
Actions:
- Apply to 40–60 roles over 4 weeks, weighted toward mid-size companies and teams that are growing
- Customize resume for 3 role categories: pure data science, ML engineering-adjacent, analytics-heavy DS
- Practice SQL and Python coding challenges daily (20–30 minutes on StrataScratch or LeetCode)
- Prepare for case study interviews: "How would you predict X?" structure — clarify the problem, propose an approach, discuss evaluation, address deployment
- Network strategically: attend local meetups, engage on LinkedIn, reach out to data scientists at target companies
Certifications complement the portfolio but don't replace it. For which ones are worth your time and money, see Best Data Science Certifications.
Portfolio beats certifications for getting hired in data science. Three polished projects — EDA, end-to-end ML, and a specialization piece — demonstrate more applied skill than any credential alone. The job search starts in Month 11, not after everything feels "perfect." Apply to 40–60 roles over 4 weeks, customized by role category.
Not every resource is worth the time. These are the highest-signal options for each phase of the roadmap.
| Resource | Type | Best For | Cost |
|---|---|---|---|
| Kaggle micro-courses | Free courses | Python, pandas, ML basics — browser-based, no setup | Free |
| Andrew Ng's ML Specialization (Coursera) | Video course | ML theory + intuition — the gold standard for understanding algorithms | $49/month |
| fast.ai Practical Deep Learning | Free course | Deep learning — top-down approach, real projects from Day 1 | Free |
| Hands-On ML by Aurélien Géron (O'Reilly) | Book | Complete ML + DL reference — code-first, scikit-learn + TensorFlow/Keras | ~$55 |
| Python for Data Analysis by Wes McKinney (O'Reilly) | Book | pandas mastery — written by the library's creator | ~$45 |
| Build a Career in Data Science by Robinson & Nolis (Manning) | Book | Career strategy — job search, interviews, workplace skills | ~$40 |
| Khan Academy Statistics | Free course | Statistics foundation — structured, self-paced | Free |
| StrataScratch | Practice platform | Real SQL + Python interview questions from actual companies | Free tier available |
The best learning path combines free resources for foundations (Kaggle, Khan Academy, fast.ai) with one definitive reference book (Hands-On ML by Géron for ML and deep learning, Python for Data Analysis by McKinney for pandas). Paid courses are optional — Andrew Ng's Coursera specialization is the best investment if choosing one.
- 01Months 1–3: Python + statistics foundation — build fluency in pandas, NumPy, probability, and hypothesis testing. Deliverable: EDA notebook with statistical analysis on a real dataset
- 02Months 4–6: ML + SQL — learn scikit-learn, supervised/unsupervised learning, feature engineering, and SQL for data access. Deliverable: end-to-end ML project with cross-validation and business interpretation
- 03Months 7–9: Deep learning + specialization — learn PyTorch or TensorFlow, pick NLP, CV, or time series. Deliverable: specialization project with trained neural network
- 04Months 10–12: Portfolio + job search — polish 3 projects, deploy a model, build resume, apply to 40–60 roles. The portfolio is the product — certifications are supporting evidence
- 05Total timeline: 12 months at 15–20 hours/week, or 6–9 months full-time. Career changers with math/CS backgrounds can compress to 6–9 months part-time
Can I follow this roadmap while working full-time?
Yes. The roadmap assumes 15–20 hours per week, which is manageable alongside a full-time job — typically 2–3 hours on weekday evenings and 5–6 hours on weekends. The 12-month timeline accounts for part-time study. Consistency matters more than intensity: 15 hours every week beats 40 hours one week followed by zero the next.
Do I need a master's degree to become a data scientist?
Not strictly, but it helps. Roughly 60% of data science job postings list a master's or PhD as preferred. The portfolio-first approach in this roadmap is designed to compensate: 3 end-to-end projects with a deployed model demonstrate applied skill that a degree alone does not. Many companies — especially startups and mid-size tech firms — hire based on demonstrated ability over credentials.
Should I learn R or Python?
Python. The Kaggle Survey consistently shows 87%+ of data scientists use Python as their primary language. R remains strong in academic research and biostatistics, but Python dominates in industry. Learning R after Python is straightforward if a future role requires it — but starting with Python maximizes job market access.
What if I already have a strong math background?
Skip or accelerate the statistics portions of Months 2–3 and invest that time in deeper ML theory or earlier specialization. A strong math background (calculus, linear algebra, probability theory) is the single biggest accelerator for this roadmap — it compresses the 12-month timeline to 6–9 months because the statistical foundation is already in place.
How important are Kaggle competitions for getting hired?
Useful but not essential. A top 10% finish on a relevant Kaggle competition is a strong portfolio signal. But most hiring managers care more about end-to-end projects — problem framing, data cleaning, feature engineering, model evaluation, and business interpretation — than competition leaderboard rankings. Kaggle competitions optimize for prediction accuracy; real data science jobs require the full pipeline.
Prepared by Careery Team
Researching Job Market & Building AI Tools for careerists · since December 2020
- 01Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow — Aurélien Géron (2022 (3rd edition))
- 02Python for Data Analysis: Data Wrangling with pandas, NumPy, and Jupyter — Wes McKinney (2022 (3rd edition))
- 03Build a Career in Data Science — Emily Robinson, Jacqueline Nolis (2020)
- 04Occupational Outlook Handbook: Data Scientists — Bureau of Labor Statistics (2025)
- 05State of Data Science and Machine Learning (Kaggle Survey) — Kaggle (2022)