How to Prepare for Data Scientist Exams: Practice Questions & Study Guide

A practical guide to data scientist exam prep — what topics are tested, practice questions for each area, and how to measure your readiness before test day.

Examifyr·Sep 2026·8 min read

Data scientist technical assessments test a surprisingly consistent set of topics — NumPy, Pandas, statistics, and machine learning fundamentals. This guide walks you through what to study, in what order, and includes a practice question for each area so you can test yourself as you go.

What Data Scientist Exams Actually Test

Whether it is a university examination, a take-home assessment, or a live coding interview, data scientist exams tend to cluster around six areas:

NumPy — array manipulation, broadcasting, indexing
Pandas — DataFrames, filtering, GroupBy, merge
Statistics — distributions, hypothesis testing, p-values
Machine learning fundamentals — supervised vs. unsupervised, bias-variance tradeoff
Model evaluation — precision, recall, F1, ROC-AUC
Feature engineering — encoding, scaling, handling missing data

The good news: these six areas have limited scope. You can cover everything examiners test in a focused study week if you know what to prioritise.

1. NumPy — Start Here

NumPy questions appear on almost every data science exam because it underpins everything else. The three concepts that appear most often:

Broadcasting — how NumPy operates on arrays of different shapes
Boolean indexing — filtering with a[a > 0]
Vectorised operations — why loops are replaced with np.sum, np.mean

import numpy as np

a = np.array([1, -2, 3, -4, 5])

# Boolean indexing — keep only positives
a[a > 0]          # [1, 3, 5]

# Broadcasting — add a scalar to every element
a + 10            # [11, 8, 13, 6, 15]

# Aggregation
a.mean()          # 0.6
np.abs(a).sum()   # 15

Practice question: Given a = np.array([[1, 2], [3, 4]]) and b = np.array([10, 20]), what does a + b produce? Answer: [[11, 22], [13, 24]] — b is broadcast across each row.

Study the full NumPy topic in the NumPy Basics guide.

2. Pandas — The Most-Tested Library

Pandas is the most common subject of data scientist technical assessments. The topics examiners reach for most:

.loc vs .iloc — label-based vs. integer position-based selection
Boolean filtering with & and | (not and / or)
groupby().agg() — splitting data and aggregating results
merge() — equivalent to SQL JOINs

import pandas as pd

df = pd.DataFrame({
    'dept':   ['Eng', 'Eng', 'Product', 'Design'],
    'salary': [70000, 85000, 90000, 75000],
})

# GroupBy — average salary per department
df.groupby('dept')['salary'].mean()

# Filter — engineers earning above 80k
df[(df['dept'] == 'Eng') & (df['salary'] > 80000)]

Practice question: What is the difference between df.loc[1:3] and df.iloc[1:3]? Answer: .loc uses index labels (includes row 3); .iloc uses integer positions (excludes row 3 — Python-style slicing).

Study the full topic in the Pandas Basics guide.

3. Statistics — The Theory Layer

Statistics questions test conceptual understanding, not just code. The most common traps:

P-value misconception — a p-value is NOT the probability that the null hypothesis is true. It is the probability of observing this data assuming the null is true.
Mean vs. median — mean is sensitive to outliers; median is robust. When they diverge, suspect skew or outliers.
Distributions — know Normal (68-95-99.7 rule), Binomial, and Poisson.

Practice question: A hypothesis test gives p = 0.03. What does this mean? Answer: If the null hypothesis were true, there is a 3% chance of observing data this extreme or more. It does not mean there is a 3% chance the null hypothesis is true.

Study the full topic in the Statistics Fundamentals guide.

4. Machine Learning — Concepts First, Code Second

ML exam questions are mostly conceptual. The concepts that appear most:

Overfitting vs. underfitting — overfitting: low train error, high test error; underfitting: high both
Bias-variance tradeoff — high bias = underfitting, high variance = overfitting
Train / validation / test split — the test set must only be touched once at the very end
Supervised vs. unsupervised — labelled data vs. finding structure without labels

Practice question: Your model achieves 99% training accuracy and 62% test accuracy. What is the problem, and name two ways to fix it? Answer: Overfitting. Fixes: add more training data, apply regularisation (L1/L2, dropout), or use a simpler model.

Study the full topic in the Machine Learning Overview guide.

5. Model Evaluation — Know Your Metrics

The most common model evaluation question: “When would you use precision vs. recall?”

Precision — use when false positives are costly (spam filter, fraud alerts)
Recall — use when false negatives are costly (cancer screening, fraud prevention)
F1 — use when you need to balance both
ROC-AUC — threshold-independent; useful for imbalanced datasets

Practice question: A cancer detection model has 98% accuracy on a dataset where 2% of patients have cancer. Is accuracy a good metric here? Answer: No — a model that predicts “no cancer” for everyone would also achieve 98% accuracy. Use recall, since missing a cancer case (false negative) is far more costly than a false alarm.

Study the full topic in the Model Evaluation guide.

6. Feature Engineering — Often the Difference-Maker

Feature engineering questions test whether you understand why certain preprocessing steps are necessary, not just how to apply them.

Scale features? Yes for KNN, SVM, neural networks, logistic regression. No for tree-based models.
Fit scaler on training data only — never fit on the test set (data leakage).
Categorical encoding — label encoding for ordinal; one-hot for nominal (unordered).

Study the full topic in the Feature Engineering guide.

Suggested Study Order

Read each guide section above (total: ~35 minutes)
Answer the practice question at the end of each section without looking
Take the free data science readiness quiz to get a topic-by-topic score
Return to the guides for any topic where you scored below 70%
Re-take the quiz — aim for 80%+ before your exam

How to Spot Your Weak Areas

The fastest way to identify gaps before your exam is to take the free data science practice quiz. It covers all six topic areas and gives you a readiness score (0–100) plus a topic breakdown showing exactly where to spend more time. No sign-up required — results are instant.

🎯

Think you're ready? Prove it.

Take the free Data Science readiness test. Get a score from 0–100, a topic breakdown, and your exact weak areas — in under 20 minutes.

Take the free Data Science test →

Free · No sign-up · Instant results