Data Science Roadmap 2026: From Beginner to Job-Ready

A data scientist uses data, statistics and machine learning to answer questions and make predictions: which customers will leave, how much stock to order, whether a new feature worked. The job mixes analysis, modelling and communication. Strong fundamentals in statistics, SQL and Python matter more than knowing the newest algorithm.

This roadmap takes about 8 to 12 months part time from a beginner level, or 4 to 6 months if you already know Python and SQL (skip or skim Stages 1 and 2).

How to use this roadmap

• Learn with real datasets from the start, not only tutorials.
• Explain everything in plain words. Interviewers care whether you understand when and why to use a method.
• Finish projects. Two finished, well-explained projects beat ten half-done notebooks.

The skills you need

Area	What to learn
Programming	Python, pandas, NumPy, plotting, notebooks, Git
Data	SQL, cleaning data, joining sources
Statistics	Probability, distributions, hypothesis tests, A/B testing
Machine learning	Regression, classification, clustering, trees and boosting
Evaluation	Validation, metrics, leakage, thresholds, interpretability
Production	Pipelines, deployment, monitoring, basic MLOps
Communication	Clear charts, short write-ups, recommendations

Projects that show you are ready

1. Churn or default prediction on a realistic dataset: a clear target, honest validation, a chosen threshold and a business recommendation.
2. An A/B test analysis: sample size, test, result and what you would decide.
3. A text or image project using a pre-trained model, compared with a simple baseline.

Common mistakes

1. Jumping to deep learning before mastering statistics and classic ML.
2. Leaky validation that makes models look better than they are.
3. No business context. Always state the decision your model supports and how you would measure its impact.

Next steps

Prepare for interviews with the data science interview questions and the machine learning interview questions. If you are choosing between data roles, read data engineer vs data scientist vs data analyst. Interested in building AI products instead? See the AI engineer roadmap.

The plan

1. 01

   Stage 1

   Stage 1: Python, NumPy and pandas

   4–6 weeks

   Python basics, then working with tables in pandas: loading, cleaning, filtering, grouping, joining and plotting.

   Outcome

   You can load a messy CSV, clean it and answer ten questions about it with pandas and charts.

   Start with

   • LessonPython for Data Engineering→
   • Lessonpandas for Data Engineers: Essentials and When Not to Use It→
   • LessonPython File I/O for Data Engineers: CSV, JSON, JSON Lines and Gzip→
2. 02

   Stage 2

   Stage 2: SQL

   3 weeks

   Joins, aggregation, window functions and CTEs, because most real data lives in databases and SQL is tested in most data science interviews.

   Outcome

   You can write a query that builds a training table with one row per customer from several tables.

   Start with

   • LessonSQL Fundamentals: SELECT, WHERE, DISTINCT, ORDER BY and LIMIT→
   • LessonSQL Joins: INNER, LEFT, RIGHT, FULL, Self and Cross Joins→
   • LessonSQL Window Functions: Ranking, LAG/LEAD and FIRST_VALUE→
3. 03

   Stage 3

   Stage 3: Statistics and probability

   4 weeks

   Distributions, mean and variance, sampling, confidence intervals, hypothesis tests, p-values and A/B testing.

   Outcome

   You can design an A/B test, choose a sample size and explain the result correctly.

   Start with

   • LessonDescriptive Statistics: Mean, Median, Spread and Outliers→
   • LessonProbability Distributions and the Central Limit Theorem→
   • LessonHypothesis Testing and P-Values Explained Simply→
   • LessonA/B Testing: Design, Sample Size and Analysis→
4. 04

   Stage 4

   Stage 4: Machine learning fundamentals

   6 weeks

   Supervised and unsupervised learning, linear and logistic regression, trees, random forest, gradient boosting, k-means and PCA, with scikit-learn.

   Outcome

   You can train, compare and explain three models on the same problem.

   Start with

   • LessonThe Machine Learning Workflow: From Data to a Working Model→
   • LessonLinear and Logistic Regression Explained, with Python→
   • LessonDecision Trees, Random Forests and Gradient Boosting→
   • Career guideTop 50 Machine Learning Interview Questions and Answers→
5. 05

   Stage 5

   Stage 5: Features, evaluation and avoiding leakage

   3 weeks

   Feature engineering, train/validation/test splits, cross-validation, the right metric for the problem, imbalanced data, thresholds and data leakage.

   Outcome

   You can justify your metric and show your validation score is honest.

   Start with

   • LessonEvaluation Metrics: Precision, Recall, ROC AUC and Thresholds→
   • LessonOverfitting, Cross-Validation and Data Leakage→
6. 06

   Stage 6

   Stage 6: Deep learning and LLM basics

   4–6 weeks

   Neural networks, training loops in PyTorch, transfer learning for text and images, and how large language models work.

   Outcome

   You can fine-tune a small pre-trained model and compare it with a classic baseline.

   Start with

   • LessonNeural Networks Explained Simply, with a Network You Can Run→
   • LessonHow Neural Networks Learn: Loss, Gradient Descent and Backpropagation→
   • LessonCNNs, RNNs and Transformers: The Main Neural Network Designs→
   • LessonPyTorch Basics: Tensors, Models and the Training Loop→
7. 07

   Stage 7

   Stage 7: From notebook to production

   3 weeks

   Pipelines, experiment tracking, model versioning, batch and online prediction, monitoring for drift, and Docker.

   Outcome

   Your model runs as a scheduled batch job or a small API, with logged predictions.

   Start with

   • LessonWhat Is MLOps? The Machine Learning Lifecycle in Production→
   • LessonExperiment Tracking and Model Registries, with Python→
   • LessonModel Serving: Batch vs Online Predictions→
   • LessonModel Monitoring and Data Drift: Keeping Models Healthy→
8. 08

   Stage 8

   Stage 8: Portfolio and interviews

   3–4 weeks

   Two or three end-to-end projects with clear business questions and results, plus practice on ML, statistics, SQL and case questions.

   Outcome

   You can explain each project's problem, data, model, evaluation and impact in five minutes.

   Start with

   • Career guideTop 50 Machine Learning Interview Questions and Answers→
   • Career guideData Engineer vs Data Scientist vs Data Analyst: What Each One Does→