Data Science Internship Interview Prep: Singapore Guide

Data Science Internship Interview Prep: Singapore Guide

Data science is one of the fastest-growing internship categories in Singapore. GovTech, Sea Group (Shopee), Grab, DBS, Standard Chartered, and dozens of funded startups hire data science interns. The interview covers four distinct areas: SQL, statistics, machine learning concepts, and case studies. This guide prepares you for all four.

The Singapore Data Science Interview Structure

Section 1: SQL

SQL is tested at every company. Proficiency up to and including window functions is expected.

Core skills to master:

• SELECT, WHERE, GROUP BY, HAVING, ORDER BY — Foundational. Must be automatic.
• JOINs (INNER, LEFT, RIGHT, FULL OUTER) — Know when to use each and how NULLs behave
• Subqueries and CTEs — CTEs (WITH statements) are preferred for readability; subqueries are sometimes faster
• Window functions: ROW_NUMBER(), RANK(), DENSE_RANK(), LAG(), LEAD(), SUM() OVER(PARTITION BY...) — These are the most commonly tested advanced SQL concepts

Common question patterns:

"Find the top 3 products by revenue for each region." → Use RANK() or ROW_NUMBER() with PARTITION BY.

"Find users who made a purchase in January but not in February." → LEFT JOIN or NOT IN / NOT EXISTS pattern.

"Calculate the 7-day rolling average of daily orders." → SUM() OVER (ORDER BY date ROWS BETWEEN 6 PRECEDING AND CURRENT ROW)

Practice resources: LeetCode SQL section (Medium difficulty), Mode Analytics practice problems, StrataScratch.

Section 2: Statistics

Probability fundamentals:

• Bayes' theorem and conditional probability
• Probability distributions: Normal, Bernoulli, Binomial, Poisson, Exponential
• Central Limit Theorem: Why is it important? (It allows us to make inferences about population parameters using sample statistics)
• Law of Large Numbers

Hypothesis testing:

• Null hypothesis (H0) vs alternative hypothesis (H1)
• Type I error (false positive, α) and Type II error (false negative, β)
• p-value: The probability of observing results at least as extreme as the data, assuming H0 is true. Does NOT tell you the probability that H0 is true.
• t-test (comparing means), chi-square test (categorical variables), ANOVA (multiple groups)

Common interview question: "You run an A/B test and get p=0.04 with α=0.05. What do you conclude?" Answer: You reject H0 — there is statistically significant evidence of a difference. However, always add: "Statistical significance does not equal practical significance — I'd also look at the effect size and business impact."

A/B testing: Expect questions like: "How would you design an A/B test to evaluate a new feature on Grab's app?" Structure your answer: Define the metric (primary KPI + guardrails), determine sample size (using power analysis: need minimum detectable effect, α, and power 1-β), determine random assignment (user-level or session-level?), set the test duration, and plan the analysis.

Section 3: Machine Learning Concepts

For most internship interviews, conceptual understanding is more important than implementation code.

Supervised learning:

• Linear regression: assumptions (linearity, homoscedasticity, independence), interpretation of coefficients, R²
• Logistic regression: logit function, odds ratio, binary classification
• Decision trees: splitting criteria (Gini impurity, entropy), overfitting, pruning
• Random forests: ensemble method, bagging, feature importance
• Gradient boosting (XGBoost, LightGBM): boosting vs bagging, overfitting control

Unsupervised learning:

• K-means clustering: how it works, choice of K (elbow method, silhouette score)
• PCA: dimensionality reduction, explained variance ratio

Model evaluation:

• Accuracy, Precision, Recall, F1 Score — and when to prioritise each
• ROC-AUC curve — what it means, when it is misleading (class imbalance)
• Confusion matrix — be able to draw and interpret one

Common interview question: "Your model has 99% accuracy but your boss is unhappy. Why might this be?" Answer: Class imbalance. If 99% of transactions are legitimate and 1% are fraudulent, a model that predicts "legitimate" for everything achieves 99% accuracy while catching zero fraud.

Section 4: Case Study Questions

Data science case studies test your ability to approach a business problem with data thinking.

Example: "Shopee's conversion rate dropped 15% last week. How would you investigate?"

Strong answer:

1. Clarify the metric: Is this across all platforms (iOS, Android, desktop)? All geographies? All product categories?
2. Check for data issues: Is the data collection pipeline functioning correctly?
3. Segment: Break down by device, geography, traffic source, user cohort (new vs returning). Where is the drop most concentrated?
4. Hypotheses: UI change? Payment gateway issue? Price increase? Seasonal effect? Competitor promotion?
5. Test: Which hypothesis can you test most quickly with available data?
6. Action: What would you recommend based on the most likely cause?

GovTech DS Interview Specifics

GovTech data science interviews emphasise:

• Python (pandas, scikit-learn) over R
• Data visualisation (seaborn, plotly, or Tableau)
• Public sector use cases: fraud detection, resource allocation, citizen service personalisation
• Data privacy awareness: PDPA compliance, data minimisation, anonymisation techniques

Grab / Sea DS Interview Specifics

Grab focuses on geospatial data analysis (ETA models, surge pricing, driver allocation) and recommendation systems. Sea/Shopee focuses on user behaviour modelling, GMV forecasting, and pricing optimisation.

For both, be ready to discuss a real personal project in Python or SQL. Candidates who can walk through a live notebook or GitHub project during a technical interview consistently outperform those who can only answer abstract questions.