130+ Data Science Technical & Scenario Interview Questions

Technical data science interviews assess far more than textbook definitions — top tech firms evaluate your ability to think under pressure, reason through mathematical trade-offs, write optimized SQL on the fly, and defend architectural decisions. Below is a curated selection from our master question bank.

Track 1: Statistical Foundations & Probability

Q1: Explain the Central Limit Theorem and its practical application.

Answer: The Central Limit Theorem (CLT) states that given a population with arbitrary distribution, finite mean \(\mu\), and finite variance \(\sigma^2\), the sampling distribution of the sample mean approaches a Normal Distribution \(\mathcal{N}(\mu, \sigma^2/n)\) as sample size \(n \ge 30\). In practice, it justifies using parametric z-tests and t-tests on skewed real-world metrics like transaction spend or session times.

Q2: A product manager checked an A/B test daily and stopped it on Day 4 when p = 0.03. Why is this invalid?

Answer: This is the classic Continuous Peeking Problem (multi-testing fallacy). Repeatedly evaluating significance before reaching predetermined sample sizes inflates the Type I error rate from 5% to over 30%! Furthermore, stopping on Day 4 misses weekly cyclical patterns (weekend vs. weekday behavior). The experiment must run for its full pre-calculated sample size and duration.

Track 2: SQL & Data Manipulation Challenges

Q3: Write a SQL query to find the 2nd highest salary in an employee table without using LIMIT.

-- Robust solution using DENSE_RANK to handle salary ties gracefully
WITH RankedSalaries AS (
    SELECT salary, DENSE_RANK() OVER (ORDER BY salary DESC) as rk
    FROM employees
)
SELECT salary FROM RankedSalaries WHERE rk = 2;

Q4: How does NULL behave in SQL operations?

Answer: In ANSI SQL, NULL represents an unknown state, not zero or an empty string. Any arithmetic or comparison with NULL evaluates to NULL (Three-Valued Logic: True, False, Unknown). NULL = NULL evaluates to UNKNOWN, which is why IS NULL and IS NOT NULL are mandatory syntax. COUNT(*) includes NULL rows, whereas COUNT(col) ignores them.

Track 3: Machine Learning & System Architecture

Q5: What is the difference between Data Drift and Concept Drift?

Answer: Data Drift (Covariate Shift) occurs when the distribution of input features changes: \(P(X)_{\text{prod}} \neq P(X)_{\text{train}}\), while the relationship between features and target remains constant (e.g., users getting younger). Concept Drift occurs when the relationship between features and target changes: \(P(Y|X)_{\text{prod}} \neq P(Y|X)_{\text{train}}\) (e.g., shopping habits shifting abruptly during a pandemic).

Practice Tip

When interviewing for senior roles, always contextualize answers with engineering trade-offs: mention latency, memory footprint, data leakage risks, and how you would automate monitoring in production.

Frequently Asked Questions (FAQ) - Student & Interview Review

Q1. What are the common rounds in a top-tier Data Science interview?

A standard tier-1 data science interview process includes: 1. Recruiter Screen (background & portfolio), 2. Technical Coding Screen (SQL & Python live coding), 3. Statistical & ML Deep-Dive (algorithms, derivations, metric trade-offs), 4. Business Case / Product Sense (A/B testing, metric design), and 5. Behavioral / Culture Leadership round.

Q2. How should I structure answers to open-ended product scenario questions?

Use the structured STAR-D framework: Situation (business context), Target KPI (clear mathematical metric), Action (data acquisition, modeling approach), Results (quantified business impact), and Defensibility (risks, edge cases, drift monitoring).

Q3. What is the single most common SQL mistake in interviews?

Failing to handle NULL values correctly in joins or aggregations (e.g., forgetting that NULL = NULL evaluates to UNKNOWN in SQL, or not using NULLIF to prevent division-by-zero errors in percentage calculations).

Q4. How do interviewers test deep statistical understanding?

Interviewers often probe edge cases: 'Why does multicollinearity not hurt tree models?', 'How does p-hacking occur?', 'Can a model have 99% accuracy and zero business value?', and 'Explain p-value to a non-technical marketing executive'.

Q5. What questions should I ask the interviewer at the end?

Ask strategic engineering questions: 'How does your team handle model monitoring and data drift in production?', 'What is the cadence of model retraining?', and 'How do data scientists collaborate with core backend software engineering teams?'

Explore Related Data Science Tutorials & Next Steps

Broaden your statistical analysis, Python data wrangling, and production engineering expertise:

BCA & CS University Viva Guide

Review academic exam questions, viva defenses, and oral theory.

Master Data Science Cheat Sheet

Quick revision of essential formulas, distributions, and code syntax.

Data Science Career, Resume & Freelancing

Turn interview readiness into lucrative job offers and consulting contracts.

Hypothesis Testing & A/B Testing

Review product sense and statistical experiment questions.