Ce parcours n’est pas encore traduit en français : son contenu s’affiche en anglais.
Data Scientist
Use data and statistics to predict outcomes and find patterns
Pas sûr(e) que ce soit pour vous ? Trouver ma voie →
0 missions sur 9 faites
Ce métier est-il fait pour vous ?
Deux minutes avant d’y consacrer des mois
Cochez ce qui vous ressemble
Cochez celles qui sont vraies pour vous.
Les compétences que vous en tirerez
Les emplois auxquels il mène
- Data Scientist
- ML Engineer
- Research Scientist
- Applied Scientist
Avant de commencer
- ✓ Some programming experience
- ✓ Comfort with math & statistics
9 missions jusqu’à Data Scientist
4 mondes, 9 missions. Touchez un cercle pour l’ouvrir ; celui qui brille est le suivant.
Don't try to learn machine learning without the math. It's essentially applied statistics, calculus, and linear algebra under the hood.
Python is the lingua franca of Data Science. You must learn how to acquire, clean, and visualize data using the core Python data stack.
Move beyond exploring data to actually predicting outcomes. This is the heart of what separates a Data Scientist from a Data Analyst.
Graduate beyond tabular data into unstructured data (text, images) using neural networks, and learn to package your models for the real world.
En savoir plus sur ce métier
Le métier, le salaire là où vous vivez, les questions fréquentes, et où ce métier peut vous mener ensuite.
Le métier
Ce que vous feriez
- Identify business problems that data analysis can help solve
- Collect, organise and analyse data
- Create, validate, test and update algorithms and models
- Use data visualisation software to present findings
- Make business recommendations to stakeholders based on the analysis
Où vous travailleriez
- Publishing, broadcasting and content providers (about 12% of US data scientists)
- Insurance companies (about 10% in the US)
- Computer systems design firms (about 10% in the US)
- Company head offices (about 10% in the US)
- Banks and other lenders (about 6% in the US)
Toutes les étapes en détail
Le texte complet de chaque étape, à lire ou à imprimer. Avancez étape par étape dans le parcours ci-dessus.
Math & Statistics Foundation 2 étapes
Master Linear Algebra & Calculus
To understand how algorithms like gradient descent or PCA work, you must grasp matrices, vectors, and derivatives. You don't need a PhD, but the intuition is mandatory.
Tâche concrète: Watch the 'Essence of Linear Algebra' and 'Essence of Calculus' series by 3Blue1Brown on YouTube. Then, manually write out the matrix dimensions and dot product operations required to pass a 3-feature input through a simple linear equation (Y = wX + b) on paper.
Pourquoi c’est important: Without linear algebra, you won't understand how neural networks pass data. Without calculus, you won't understand how models learn and optimize their weights. You'll be blind to what the libraries are doing.
Liste de contrôle
- Understand vectors, matrices, and dot products
- Learn matrix multiplication and vector spaces
- Understand eigenvalues and eigenvectors (crucial for dimensionality reduction)
- Review differential calculus: derivatives and partial derivatives
- Understand the concept of gradients and the chain rule (backpropagation)
- Practice calculating simple derivatives by hand to cement the concepts
Probability & Statistics
Data is noisy. Statistics gives you the tools to separate signal from noise, prove your findings rigorously, and understand the certainty of a prediction.
Tâche concrète: Find a Kaggle dataset regarding e-commerce conversion rates. Formulate a null hypothesis (e.g., 'A new website layout does not increase conversion'). Run a T-test in Excel or a basic Python script to determine the p-value. Document whether you reject or fail to reject the null hypothesis, and explain what a Type I vs Type II error would mean in this scenario.
Pourquoi c’est important: Businesses will ask you 'Is this feature working?'. If you don't understand statistical significance, you will cost the company millions by endorsing random variance as real success.
Liste de contrôle
- Learn descriptive statistics: Mean, Median, Mode, Variance, Standard Deviation
- Understand probability distributions (Normal, Binomial, Poisson)
- Grasp the Central Limit Theorem and why it is foundational to data science
- Study inferential statistics: Hypothesis testing, p-values, and statistical significance
- Understand A/B testing frameworks and how to calculate sample sizes
- Learn Bayes' Theorem and Bayesian vs. Frequentist approaches
Python & Data Wrangling 3 étapes
Python Programming & SQL
You need to know how to write clean code and how to extract data from corporate databases using SQL.
Tâche concrète: Connect to a public API (like OpenWeatherMap or a Crypto price API) and write a Python script that fetches JSON data, parses it, and writes the structured data into a local SQLite database using SQL INSERT statements.
Pourquoi c’est important: Data doesn't magically appear in CSV files. As a Data Scientist, you will spend 60-80% of your time just extracting data using SQL, and writing Python scripts to pipeline it.
Liste de contrôle
- Master Python basics: Lists, Dictionaries, Functions, Classes, and Error Handling
- Learn how to interact with APIs using the `requests` library
- Understand list comprehensions and lambda functions
- Write complex SQL queries (Window functions, CTEs, complex JOINs) to extract datasets
- Learn to use Git and GitHub to version control your Jupyter Notebooks and scripts
Data Manipulation (Pandas & NumPy)
Pandas is the workhorse. You must be able to filter, aggregate, and reshape tabular data effortlessly.
Tâche concrète: Download a dirty real-world dataset (e.g., Titanic or Housing Prices from Kaggle). Write a Jupyter Notebook that loads the data, identifies missing values and imputes them based on class medians, creates 3 new computed features (feature engineering), and exports a clean CSV. Do not use loops; rely entirely on Pandas vectorized methods.
Pourquoi c’est important: Raw data is dirty, missing, and incorrectly formatted. If you cannot wrangle data efficiently with Pandas, you cannot feed it into your machine learning models.
Liste de contrôle
- Master NumPy arrays, vectorization, and broadcasting for fast math operations
- Learn Pandas DataFrames and Series structures
- Perform filtering, sorting, handling missing values (NaN), and dropping duplicates
- Master GroupBy operations, pivot tables, and multi-index aggregations
- Learn to merge, join, and concatenate multiple data sources
- Perform time-series resampling and windowing
Data Visualization & EDA
Exploratory Data Analysis (EDA) is where you actually find the insights. Visualization helps communicate findings to non-technical stakeholders.
Tâche concrète: Using the cleaned data from your previous task, create a comprehensive EDA report. Generate a correlation heatmap of numerical variables. Plot distribution graphs for the target variable. Write Markdown cells beneath each chart explaining the business insight (e.g., 'Feature X has a high positive correlation with Feature Y').
Pourquoi c’est important: A model is useless if you can't convince the business to act on it. Visualizations are how you translate math into actionable business strategies.
Liste de contrôle
- Learn Matplotlib for basic line, bar, scatter, and histogram plots
- Learn Seaborn for statistical plots (box plots, violin plots, correlation heatmaps)
- Learn Plotly or Altair for interactive dashboard-style visualizations
- Understand data storytelling: choosing the right chart for the right narrative
- Know how to identify outliers and skewness visually
Machine Learning Core 2 étapes
Supervised Learning Algorithms
Learn algorithms where you have known historical outcomes to train against (labeled data).
Tâche concrète: Take the Kaggle 'Predict Future Sales' or 'Titanic' dataset. Train a Logistic Regression model and a Random Forest model. Predict outcomes on the test set. Compare which performed better and hypothesize why the tree-based model captured relationships the linear model missed.
Pourquoi c’est important: Supervised learning solves 90% of business prediction problems. Random Forests and XGBoost are the undisputed champions of tabular data prediction.
Liste de contrôle
- Understand the Scikit-Learn API structure (fit, predict, transform)
- Learn Linear Regression (for continuous values) and Logistic Regression (for binary classification)
- Study Decision Trees and ensemble methods (Random Forest, Gradient Boosting like XGBoost)
- Learn Support Vector Machines (SVM) and K-Nearest Neighbors (KNN)
- Understand the Bias-Variance tradeoff and overfitting vs underfitting
- Implement L1 (Lasso) and L2 (Ridge) regularization to prevent overfitting
Unsupervised Learning & Evaluation
Learn algorithms to find patterns in unlabeled data, and master how to prove your model is actually good.
Tâche concrète: Use K-Means to cluster a dataset of customer shopping behaviors into distinct 'segments' without knowing them beforehand. Provide a business profile for each cluster (e.g., 'High spend/high frequency vs low spend/low frequency'). Separately, take your Random Forest from the previous task, evaluate its F1 Score (not just accuracy), and tune its max-depth using Cross-Validation to improve performance.
Pourquoi c’est important: An accuracy of 99% on an imbalanced dataset (like fraud detection) is completely fake. You must know Precision, Recall, and cross-validation to not embarrass yourself in production.
Liste de contrôle
- Learn K-Means Clustering and Hierarchical Clustering for grouping similar data
- Understand Principal Component Analysis (PCA) for reducing feature dimensionality
- Master classification metrics: Accuracy, Precision, Recall, F1 Score, and ROC-AUC
- Master regression metrics: RMSE, MAE, R-Squared
- Understand how to avoid data leakage during training
- Implement k-fold Cross Validation and Hyperparameter tuning (GridSearchCV / RandomizedSearchCV)
Deep Learning, NLP & Deployment 2 étapes
Deep Learning & NLP Foundations
Neural Networks drive modern AI. Learn how to train models using PyTorch or TensorFlow, focusing on unstructured data like text or images.
Tâche concrète: Use the Hugging Face `transformers` library to download a pre-trained sentiment analysis model (e.g., DistilBERT). Write a script that takes a list of movie reviews from IMDB dataset, passes them through the model, and outputs a CSV of the positive/negative sentiment for each.
Pourquoi c’est important: While standard business data uses XGBoost, cutting-edge products rely on Deep Learning and LLMs (Large Language Models). NLP experience is arguably the most highly sought-after skill right now.
Liste de contrôle
- Understand Perceptrons, Artificial Neural Networks (ANN), and activation functions
- Learn Convolutional Neural Networks (CNNs) for image processing
- Learn basics of Natural Language Processing (NLP): Regex, Tokenization, Stemming/Lemmatizing, TF-IDF
- Understand Word Embeddings (Word2Vec) and foundational Recurrent Neural Networks (RNN/LSTM)
- Familiarize yourself with the Hugging Face transformer ecosystem (BERT, GPT variants) for modern NLP tasks
- Implement deep learning using PyTorch (preferred) or TensorFlow/Keras
Model Deployment (MLOps Basics)
A model in a notebook generates zero revenue. You must learn how to wrap it in an API so developers can integrate it into software.
Tâche concrète: Train a simple classification model. Save it as a `.pkl` file. Build a FastAPI app with a `/predict` endpoint that loads the `.pkl` model, accepts JSON input features, and returns the model's prediction. Test the API locally using Postman or cURL.
Pourquoi c’est important: The 'Notebook to Production' gap kills 80% of data science projects. If you can deploy your own models as an API, you instantly become 2x more valuable to an engineering team.
Liste de contrôle
- Learn how to serialize/save your trained models using Pickle or Joblib
- Build an API wrapper for your model using FastAPI or Flask
- Learn basic Docker commands to containerize your API environment
- Understand how to host the containerized API on a cloud provider (AWS EC2, Heroku, or GCP Cloud Run)
- Integrate MLflow (or similar) to log experiments, metrics, and parameters
Questions fréquentes : Data Scientist
Des réponses claires sur la durée, les prérequis, les salaires et les méthodes d’étude.
Combien de temps faut-il pour devenir Data Scientist ?
Ce parcours estime 8–18 months. Il compte 4 phases et 9 étapes, de « Math & Statistics Foundation » à « Deep Learning, NLP & Deployment ».
Comment devenir Data Scientist ?
Suivez les 4 phases dans l’ordre, en commençant par la phase 1, « Math & Statistics Foundation ». Votre première tâche : Watch the 'Essence of Linear Algebra' and 'Essence of Calculus' series by 3Blue1Brown on YouTube. Then, manually write out the matrix dimensions and dot product operations required to pass a 3-feature input through a…
Combien gagne un Data Scientist ?
Le salaire dépend du pays. L’encadré salaire de cette page indique l’estimation de DraftMyCareer, en monnaie locale, pour le pays choisi dans le menu.
De quoi ai-je besoin avant de commencer ?
Ce parcours indique : Some programming experience; Comfort with math & statistics.
De quelles compétences un Data Scientist a-t-il besoin ?
Ce parcours couvre : Python, Machine Learning, Statistics, SQL, Data visualization, Experimentation.
Changer de métier : passerelles vers ce poste et depuis ce poste
Voir toutes les reconversions →Venir à ce poste depuis un autre métier
- Depuis: Data Analyst
Transition from backward-looking reporting to forward-looking predictive modeling and algorithmic decision-making.
⏱ 4–7 months🧭 ModerateCe que vous gardezSQL & Data Wrangling · Business Intelligence (Tableau/PowerBI) · Data Storytelling
Voir le plan d’action
- Month 1–2: Advanced Statistical Inference Master probability distributions, regression diagnostics, and experimental A/B testing.
- Month 3–4: Core Machine Learning Build predictive supervised and unsupervised models with feature selection in Python.
- Month 5–7: Real-world Modeling Projects Develop end-to-end churn prediction, recommendation systems, and model interpretability (SHAP).
- Depuis: Financial Analyst
Combine financial domain acumen with state-of-the-art machine learning in FinTech and hedge funds.
⏱ 5–8 months🧭 ModerateCe que vous gardezQuantitative analysis · Financial modeling in Excel · Risk assessment & forecasting
Voir le plan d’action
- Month 1–2: Python Programming & SQL Transition spreadsheet models to Python scripts, pandas DataFrames, and relational databases.
- Month 3–4: Statistical & Machine Learning Implement regression, tree ensembles (XGBoost), and time-series forecasting (ARIMA/Prophet).
- Month 5–8: Quantitative Portfolio Projects Build algorithmic trading simulations, credit risk scoring models, and fraud detection systems.