Data Science · Beginner Terms
Data Scientist Terms Explained for Absolute Beginners
Quick summary — data scientist terms
The data scientist terms beginners need most follow a project from start to finish. Data has rows, features and a label to predict; models learn on a training set and are checked on a test set; overfitting means memorising instead of learning; and metrics like precision and recall show how useful predictions are.
In this guide you will learn:
- Data terms — dataset, feature, label, outlier and more.
- Training terms — model, training set, overfitting, hyperparameter.
- Evaluation terms — accuracy, precision, recall, confusion matrix.
- A personal term log — template to keep learning.
- Tips — learning vocabulary through small projects.
SECTION 01Data terms: what you start with
Every data science project begins with a dataset — usually a table. Each row is one example, such as one customer or one flat. Each column describes something about it. The columns you use to make a prediction are called features, and the column you want to predict is the label or target. Understanding these words makes every tutorial easier to follow.
| Term | Plain meaning | Everyday example | Where you see it |
|---|---|---|---|
| Feature | An input column used for prediction | Area and location of a flat | X in code |
| Label / target | The value you want to predict | The flat's monthly rent | y in code |
| Outlier | A value far from the rest | A rent ten times the average | Charts, summary stats |
| Missing value | A blank cell in the data | Unknown number of bathrooms | NaN in pandas |
import pandas as pd
flats = pd.DataFrame({
"area_sqft": [650, 900, 1200, 450, 1500],
"bedrooms": [1, 2, 3, 1, 3],
"metro_km": [0.5, 2.0, 1.2, 3.5, 0.8],
"rent": [18000, 24000, 32000, 12000, 41000],
})
X = flats[["area_sqft", "bedrooms", "metro_km"]] # features
y = flats["rent"] # label / target
print(X.shape, y.shape)import numpy as np
flats.loc[2, "bedrooms"] = np.nan # a missing value
print(flats.isna().sum()) # count missing per column
q1, q3 = flats["rent"].quantile([0.25, 0.75])
iqr = q3 - q1
outliers = flats[(flats["rent"] < q1 - 1.5 * iqr) | (flats["rent"] > q3 + 1.5 * iqr)]
print(outliers)SECTION 02Training terms: how models learn
A model is a mathematical rule that maps features to a prediction. Training means showing the model many examples so it can adjust its internal settings. To check whether it really learned, you hide some data during training — the test set — and use it only at the end. The biggest beginner trap is overfitting: a model that performs brilliantly on training data but poorly on new data.
| Term | Plain meaning | Everyday example | Where you see it |
|---|---|---|---|
| Training set | Data the model learns from | Practice questions | X_train, y_train |
| Test set | Unseen data for the final check | The real exam | X_test, y_test |
| Overfitting | Memorising training data, failing on new data | Mugging up answers without understanding | Big gap between train and test scores |
| Hyperparameter | A setting you choose before training | Depth of a decision tree | max_depth=3 |
from sklearn.model_selection import train_test_split
from sklearn.tree import DecisionTreeRegressor
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.4, random_state=0)
deep = DecisionTreeRegressor(max_depth=None, random_state=0).fit(X_train, y_train)
shallow = DecisionTreeRegressor(max_depth=2, random_state=0).fit(X_train, y_train)
print("deep train R2:", deep.score(X_train, y_train))
print("deep test R2:", deep.score(X_test, y_test))
print("shallow train R2:", shallow.score(X_train, y_train))
print("shallow test R2:", shallow.score(X_test, y_test))How to read the output above:
- A deep tree often scores close to 1.0 on training data
(it has memorised it)
- If its test score is much lower, it is overfitting
- A shallower tree may score lower on training but
closer on test, which usually generalises better
Note: with only 5 rows this is a toy example to show
the idea. Real projects need far more data.SECTION 03Evaluation terms: is the model any good?
After training, you need to measure how useful the model is. For numbers like rent, metrics such as mean absolute error tell you how far off predictions are on average. For yes/no predictions, such as "will this customer leave?", accuracy can mislead when one outcome is rare. A confusion matrix, precision and recall give a clearer picture of what kinds of mistakes the model makes.
| Term | Plain meaning | Everyday example | When it matters |
|---|---|---|---|
| Accuracy | Share of all predictions that are correct | 95 of 100 answers right | Misleading when classes are imbalanced |
| Precision | Of the predicted positives, how many were right | Of flagged emails, how many were truly spam | When false alarms are costly |
| Recall | Of the real positives, how many were found | Of sick patients, how many the test caught | When missing cases is costly |
| Confusion matrix | Table of correct and wrong predictions by class | True/false positives and negatives | Understanding error types |
from sklearn.metrics import confusion_matrix, precision_score, recall_score, accuracy_score
y_true = [1, 0, 1, 1, 0, 0, 0, 1, 0, 0] # 1 = customer left
y_pred = [1, 0, 0, 1, 0, 1, 0, 1, 0, 0]
print(confusion_matrix(y_true, y_pred)) # [[TN FP] [FN TP]]
print("accuracy :", accuracy_score(y_true, y_pred))
print("precision:", precision_score(y_true, y_pred))
print("recall :", recall_score(y_true, y_pred))from sklearn.metrics import mean_absolute_error
actual = [18000, 24000, 32000]
predicted = [19000, 22500, 30000]
mae = mean_absolute_error(actual, predicted)
print(f"On average, predictions are off by about Rs {mae:,.0f}")SECTION 04Your personal data science term log
Keep a log as you learn. Writing a term in your own words, with an example, is the fastest way to remember it.
Term:
Stage: Data / Training / Evaluation / Statistics
My definition (one sentence):
Everyday example:
Code where it appears:
Common mistake with it:Data : dataset, row, column, categorical, numerical
Training : algorithm, baseline, cross-validation,
underfitting, regularisation
Evaluation : F1 score, ROC curve, AUC, MAE, R-squared
Learn these after the 12 in this guide.Terms learned this week:
1.
2.
3.
Term I can now explain to a friend:
Term still confusing me:
Mini experiment to understand it:SECTION 05Tips for learning data science vocabulary
These habits help terms move from confusing to obvious:
- Learn in project order — data terms first, then training, then evaluation.
- Use tiny datasets — five to ten rows let you check every number by hand.
- Draw a confusion matrix — on paper, with real counts, at least once.
- Explain with daily-life examples — exams, rent, spam and medical tests work well.
- Read code with the glossary open — match each term to where it appears.
SECTION 06Interview Q&A — data science terms
Q1What is the difference between a feature and a label?
A feature is an input variable the model uses, such as a flat's area. The label is the output you want to predict, such as its rent.
Q2What is overfitting and how can you reduce it?
Overfitting is when a model learns noise and details of the training data so well that it performs poorly on new data. You can reduce it with simpler models, more data, regularisation, cross-validation and limiting settings like tree depth.
Q3Why do we keep a separate test set?
It gives an honest estimate of how the model will perform on new, unseen data. Using it during training would make the estimate overly optimistic.
Q4When would you prefer recall over precision?
When missing a real positive is more costly than a false alarm, such as screening for a disease or detecting fraud, where you want to catch as many true cases as possible.
Q5What is a hyperparameter?
A setting chosen before training that controls how the model learns, such as the maximum depth of a tree or the number of clusters, as opposed to parameters the model learns from data.
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 are the basic terms in data science?
Start with dataset, feature, label, training set, test set, model, overfitting, accuracy, precision and recall. These appear in almost every beginner tutorial and interview.
What is the difference between data science and machine learning?
Data science is the broader practice of using data to answer questions, including analysis, statistics and communication. Machine learning is a set of methods within it for building models that learn from data.
What does "training a model" mean?
It means showing a model many examples with known answers so it can adjust its internal parameters to make better predictions on new data.
What is a good first data science project for beginners?
Predicting a number such as rent or prices from a small public dataset, or a simple yes/no classification, lets you practise features, training, test sets and evaluation metrics.
Do I need to memorise all data science terms?
No. Understand the core terms deeply and learn others as you meet them in projects. Being able to explain a few terms clearly is more valuable than memorising many.
SECTION 09Related reads
Classroom & online · Noida
Learn the terms by doing the work
Our Data Science Training Course takes you through data handling, statistics, machine learning and evaluation with projects where every term in this guide is used in practice.
Talk to a counsellor for current fees- Beginner-friendly foundations
- Hands-on ML projects
- Mentor support for doubts
- Weekday & weekend batches