Beginner Guide · Data Science Vocabulary
Data Science Terms Every Beginner Should Understand
Quick summary — the essential terms
Data science has a vocabulary problem. Most beginner confusion isn't about the math — it's about words. "Feature" sounds technical, but it just means "input column." "Label" means "the answer we want to predict." "Overfitting" means "the model memorized the training data instead of learning the pattern." Once you decode the terms, tutorials and interviews start making sense. Below is the practical glossary — grouped by category, with real examples.
SECTION 01Data terms
These are the words you'll hear in every data science tutorial during the first month. All of them describe the raw material you work with.
| Term | Plain Meaning | Real Example |
|---|---|---|
| Dataset | A collection of data in table form | CSV of 10,000 customer orders |
| Row / record | One item in the dataset | One customer order |
| Column / feature | One attribute of each row | "Price", "Date", "Customer name" |
| Label / target | The thing you're trying to predict | "Will this customer churn? Yes/no" |
| Missing values | Empty cells in the data | A customer with no email filled in |
| Categorical | Values from a fixed list | "Color: Red / Green / Blue" |
| Numerical | Values that are numbers | "Age: 34", "Price: 4999" |
| Outlier | A value far from the rest | An order worth ₹10 lakh when most are under ₹5000 |
SECTION 02Modeling terms
These describe how a model learns. This is where most beginners get confused because the words sound technical.
| Term | Plain Meaning | Why It Matters |
|---|---|---|
| Model | A rule learned from data | Every prediction comes from a model |
| Algorithm | The method that builds the model | e.g. "Random Forest" is an algorithm |
| Training | Teaching the model on data | When the model learns patterns |
| Test / evaluate | Checking the model on new data | Proves it actually works |
| Overfitting | Model memorized training data | Fails on real, new data |
| Underfitting | Model too simple to learn anything | Bad on training and test data |
| Supervised learning | Learning with labels | Most problems are this type |
| Unsupervised learning | Learning without labels | Used for clustering, discovery |
Overfitting is the most important term
for beginners to understand.
Think of it like a student who
memorizes the answers to a practice test,
then fails the real exam.
In ML:
→ The "practice test" is training data
→ The "real exam" is test data
Symptoms:
• 99% accuracy on training data
• 60% accuracy on test data
• Model looks incredible — until
it meets new data
Why it happens:
• Too many features
• Too complex model
• Too little training data
How to fix it:
• Simplify the model
• Get more data
• Use regularization
• Cross-validate# Overfitting in Python
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42
)
# Overfit model (deep tree)
overfit = DecisionTreeClassifier(max_depth=None)
overfit.fit(X_train, y_train)
print("Train:", accuracy_score(y_train, overfit.predict(X_train)))
# → 1.00 (100% on training)
print("Test: ", accuracy_score(y_test, overfit.predict(X_test)))
# → 0.62 (62% on test — bad)
# Fixed model (shallow tree)
better = DecisionTreeClassifier(max_depth=4)
better.fit(X_train, y_train)
print("Train:", accuracy_score(y_train, better.predict(X_train)))
# → 0.85
print("Test: ", accuracy_score(y_test, better.predict(X_test)))
# → 0.83
# The "worse" training score is the
# better model. This is the lesson.SECTION 03Evaluation terms
These describe how good a model is. Accuracy is not enough — especially for imbalanced data.
| Term | Plain Meaning | When It Matters |
|---|---|---|
| Accuracy | % of correct predictions | Balanced data only |
| Precision | Of those predicted positive, how many were correct? | False positives are costly |
| Recall | Of actual positives, how many were found? | False negatives are costly |
| F1 score | Balance between precision and recall | Need a single number |
| Confusion matrix | Table of predictions vs. truth | See what's failing |
| RMSE / MAE | Error size for regression | Continuous predictions |
SECTION 04Practice terms
These describe how real data scientists work — beyond just training models.
| Term | Plain Meaning | Why It Matters |
|---|---|---|
| EDA | Exploratory Data Analysis | Understanding data before modeling |
| Feature engineering | Creating better columns from existing data | The biggest impact on model quality |
| Pipeline | Automated steps: load → clean → model | Reproducible workflows |
| Cross-validation | Testing on multiple splits | More reliable evaluation |
| Baseline | Simple model to beat | Is the complex model worth it? |
| Deployment | Putting model into production | Where the model actually works |
SECTION 05Bias and fairness terms
As data science moves into real products, these terms become essential. They describe what can go wrong beyond raw accuracy.
| Term | Plain Meaning | Example |
|---|---|---|
| Sampling bias | Data doesn't represent the population | Survey of only urban users |
| Historical bias | Past discrimination in the data | Hiring data missing women candidates |
| Model bias | Model performs worse for a subgroup | Face recognition failing on darker skin |
| Fairness | Ensuring equal outcomes across groups | Equal approval rates by gender |
| Explainability | Can you explain why the model decided? | Credit decisions need a reason |
SECTION 06Career terms
These come up in job descriptions and interviews. Confusing them leads to applying for the wrong roles.
| Term | What It Means | Where It's Used |
|---|---|---|
| Data Analyst | Analyzes data, builds dashboards | Business intelligence |
| Data Scientist | Builds statistical/ML models | Product, research |
| ML Engineer | Deploys and maintains ML systems | Production ML |
| Data Engineer | Builds data pipelines and infrastructure | Data platform |
| Analytics Engineer | Bridges data engineering and analysis | Modern data stack |
| MLOps | DevOps for ML systems | Production ML |
SECTION 07Test yourself — data science terms quiz
Five questions. No sign-up.
0 / 5Pick an answer to see why it is right or wrong.
SECTION 08Frequently asked questions
What is a feature in data science?
A feature is a column in your dataset that's used as an input to predict something. "Age", "Price", and "Days since last login" are features.
What's the difference between a model and an algorithm?
An algorithm is the method (e.g. Random Forest, Logistic Regression). A model is the actual rule learned from the data using that algorithm. You train a model with an algorithm.
What is overfitting?
Overfitting means the model memorized the training data instead of learning general patterns. It performs great on training data but poorly on new data.
Why is accuracy not always a good metric?
On imbalanced data, accuracy is misleading. A model that always predicts "not fraud" on a dataset that's 99% not fraud gets 99% accuracy — but catches zero fraud. Precision and recall matter more.
What is a data pipeline?
A data pipeline is an automated sequence of steps that takes raw data and produces something useful — usually "load → clean → transform → model or visualize." Pipelines make work reproducible.
SECTION 09Related reads
Classroom & online · Noida
Start your data science journey
Our Data Analytics Course covers Python, SQL, Pandas, ML basics, and real projects — explained in plain English.
₹15,500 · full programme- Python + SQL + Pandas + ML
- Real projects & portfolio
- Mock interviews & resume help
- Weekday & weekend batches