A 2024 LinkedIn Workforce Report found "data scientist" still among the top 5 roles with unfilled demand — yet most bootcamp grads struggle with the same gaps: forgetting the interquartile range formula mid-interview, blanking on when to use a t-test vs. a chi-square, or shipping a model without ever questioning data leakage. This data science cheat sheet covers the concepts you actually need to remember, organized by the real workflow, not a textbook table of contents.
Bookmark it. Use it before interviews. Use it when you're three hours into a Jupyter notebook and can't remember whether pandas.merge defaults to an inner or outer join (it's inner).
The Data Science Cheat Sheet: Full Workflow at a Glance
Every data science project follows the same backbone regardless of domain. Here's the sequence — and the most common place people get it wrong at each step.
1. Define the Problem
- Is this a classification problem (discrete output) or regression (continuous output)?
- What's the baseline? (If 95% of transactions are non-fraud, predicting "never fraud" gives 95% accuracy — worthless.)
- What does "good" look like? Define your success metric before touching data.
2. Collect & Store Data
- Structured data → relational DB (PostgreSQL, MySQL, SQLite)
- Semi-structured → JSON/Parquet in a data lake
- Unstructured → blob storage (S3, GCS)
- Key SQL you need cold:
JOIN,GROUP BY,HAVING,window functions (RANK, LAG, LEAD),CTEs
3. Explore (EDA)
- Shape:
df.shape,df.info(),df.describe() - Missing values:
df.isnull().sum() - Distributions: histograms, box plots, violin plots
- Correlations: Pearson (linear, continuous), Spearman (monotonic, ordinal)
- Class imbalance check before any classification work
4. Clean & Feature Engineer
- Imputation: mean/median for MCAR data; model-based for MAR/MNAR
- Outlier treatment: IQR fence (
Q1 - 1.5×IQRtoQ3 + 1.5×IQR), log transform, winsorization - Encoding: one-hot (low cardinality), target encoding (high cardinality), ordinal (ordered categories)
- Scaling: StandardScaler (zero mean, unit variance) for distance-based models; MinMaxScaler for neural nets
Statistics Quick Reference — The Core of the Data Science Cheat Sheet
This is the section most people skip. Don't. Interviewers at data-heavy companies test statistics harder than Python syntax.
Descriptive Statistics
| Measure | Formula / Note | When to use it |
|---|---|---|
| Mean | Σx / n | Symmetric distributions |
| Median | Middle value after sort | Skewed data, outliers present |
| Std Dev (σ) | √(Σ(x−μ)² / N) | Spread around the mean |
| IQR | Q3 − Q1 | Robust spread, outlier detection |
| Skewness | >0 = right tail, <0 = left tail | Deciding mean vs median |
Hypothesis Testing Decision Tree
- Two means, normal data: independent t-test (unequal variance → Welch's t-test)
- Two means, paired samples: paired t-test
- More than two means: one-way ANOVA → post-hoc Tukey
- Categorical vs categorical: chi-square test of independence
- Non-normal, two groups: Mann-Whitney U
- Correlation strength: p-value tells you if it's real; r² tells you how much variance it explains
p-value reminder: p < 0.05 means the result would occur by chance less than 5% of the time under the null hypothesis. It does NOT measure effect size or practical significance.
Machine Learning Model Cheat Sheet
This data science cheat sheet wouldn't be complete without a model selection guide. The right algorithm depends on your data shape, not on what's trendy.
Supervised Learning
- Linear Regression: continuous target, linear relationship, interpretability required
- Logistic Regression: binary classification, baseline before complex models
- Decision Tree: interpretable, non-linear, prone to overfitting solo
- Random Forest: bagging ensemble — reduces variance of decision trees
- Gradient Boosting (XGBoost, LightGBM): boosting ensemble — wins most tabular competitions
- SVM: high-dimensional spaces, small datasets, text classification
Unsupervised Learning
- K-Means: fast clustering, requires k upfront, sensitive to scale
- DBSCAN: finds arbitrary shapes, handles noise, no k required
- PCA: dimensionality reduction, removes multicollinearity
Evaluation Metrics — Know These Cold
- Accuracy: (TP + TN) / total — misleading on imbalanced classes
- Precision: TP / (TP + FP) — "when I predict positive, how often am I right?"
- Recall: TP / (TP + FN) — "of all actual positives, how many did I catch?"
- F1: harmonic mean of precision and recall — use when both matter
- AUC-ROC: probability a random positive ranks above a random negative
- RMSE: penalizes large errors — regression standard
- MAE: more robust to outliers than RMSE
SQL Quick Reference
SQL is the most underrated part of any data science cheat sheet. Most real-world data science work starts in a database, not a CSV.
-- Aggregate with filter on group
SELECT category, AVG(salary) AS avg_salary
FROM employees
WHERE start_date >= '2022-01-01'
GROUP BY category
HAVING COUNT(*) > 10
ORDER BY avg_salary DESC;
-- Window function: rank within group
SELECT name, department, salary,
RANK() OVER (PARTITION BY department ORDER BY salary DESC) AS dept_rank
FROM employees;
-- CTE: readable multi-step logic
WITH monthly_sales AS (
SELECT DATE_TRUNC('month', sale_date) AS month,
SUM(amount) AS total
FROM sales
GROUP BY 1
)
SELECT month, total,
LAG(total) OVER (ORDER BY month) AS prev_month
FROM monthly_sales;Top Courses to Go Deeper
This cheat sheet gives you the reference layer. These courses give you the practice layer — working through real datasets is what actually makes this knowledge stick.
Introduction to Data Analytics
A strong starting point if you're new to the field — covers the full analytics lifecycle from asking the right question to communicating results, with hands-on work that maps directly to the workflow outlined above.
Executive Data Science Specialization
Covers how data science projects are scoped and managed in real organizations — useful both for practitioners and for anyone moving into a lead or stakeholder role who needs to bridge the gap between technical teams and business decisions.
Database Design and Basic SQL in PostgreSQL
The SQL section of this cheat sheet is only useful if you can write it fluently — this course builds that fluency through PostgreSQL, which is the industry standard for analytical workloads.
Applied Plotting, Charting & Data Representation in Python
EDA skills and visualization are where most self-taught data scientists have the biggest gaps. This course is specifically about making charts that communicate insight, not just charts that exist.
COVID-19 Data Analysis Using Python
A real-world project course using a dataset most people already have context for — good for practicing the full EDA-to-insight pipeline on messy, time-series public health data.
Introduction to Data Analysis Using Microsoft Excel
Excel is still the first tool stakeholders reach for, and knowing pivot tables, VLOOKUP, and basic statistical functions cold makes you dramatically more effective in business environments where Python isn't deployed.
FAQ
What should a data science cheat sheet include?
At minimum: the end-to-end workflow (problem definition → data collection → EDA → modeling → evaluation → deployment), core statistical tests and when to use each, SQL window functions and aggregations, model selection guidelines by problem type, and evaluation metrics with their trade-offs. This page covers all of those.
How do I memorize statistics formulas for data science interviews?
Don't try to memorize them in isolation. Practice applying them — run t-tests, compute IQR, build confusion matrices from scratch — and the formulas stick naturally. Flashcard tools like Anki work well for the handful of things you still need cold (like the IQR outlier fence or p-value interpretation).
What's the difference between data science and data analytics?
Data analytics is primarily retrospective — "what happened and why?" — and relies heavily on SQL, Excel, and BI tools. Data science adds predictive and prescriptive modeling (ML, statistical inference) and typically requires programming in Python or R. In practice, the titles are used interchangeably at many companies.
Do I need to know R for data science in 2026?
Python has won the general-purpose data science battle, but R remains dominant in academic statistics, bioinformatics, clinical research, and anywhere that publication-quality statistical output matters. If you're targeting industry roles at tech companies, Python is the priority. If you're targeting research, pharma, or academic roles, R proficiency is still a differentiator.
What SQL do data scientists actually use day-to-day?
In practice: SELECT with GROUP BY, JOIN (inner and left), HAVING, CASE WHEN, window functions (ROW_NUMBER, RANK, LAG, LEAD), and CTEs. Subqueries, date functions, and COALESCE for null handling round out the daily toolkit. Most people never need recursive CTEs or advanced set operations in production work.
How long does it take to become job-ready in data science?
With focused effort and prior programming or stats exposure: 6–12 months to be competitive for junior analyst roles. For data scientist roles requiring ML and statistics depth: 12–18 months is realistic. The bottleneck is almost always project portfolio depth, not course completion counts.
Bottom Line
The concepts on this data science cheat sheet — the EDA workflow, hypothesis testing decision tree, model selection guidelines, SQL patterns, and evaluation metrics — are what hiring managers actually test and what experienced data scientists reach for day-to-day.
If you're early in your learning, start with Introduction to Data Analytics to get the workflow down, then pair it with Database Design and SQL in PostgreSQL to solidify your SQL. Those two alone will make you competitive for analyst roles. Add Applied Plotting in Python for the visualization gap most self-taught learners have, and you'll have covered the three pillars that show up in nearly every data role interview.
Come back to this cheat sheet before your next technical screen. The fundamentals don't change.