Data Scientist technical round practice: stats, ML and coding

Most Data Scientist technical rounds sample from four areas, sometimes in a single sitting. Most Data Analyst / Data Science Jobs at the senior end expect fluency across all four strands.

• Statistics and inference: hypothesis testing, p-values, confidence intervals, and why a result may be spurious.
• Probability: conditional probability, expectation, and classic brain-teasers framed around real data.
• Machine learning: bias-variance, regularisation, evaluation metrics, and when a model choice is wrong for the problem.
• Coding: SQL plus Python or R to manipulate data and implement a small algorithm.

Interviewers care whether you know when a method is appropriate, not just how to run it.

Practise these recurring shapes rather than isolated trivia.

'Your A/B test shows a 2% lift, p=0.06. What do you do?'

They want you to discuss power, sample size, practical significance and the risk of peeking, not just recite the threshold.

'Explain the bias-variance trade-off with a concrete example.'

Strong candidates tie it to a real model and to regularisation, not textbook definitions.

'Choose an evaluation metric for a fraud model.'

This probes whether you reason about class imbalance, precision-recall, and the cost of false negatives.

Question: 'You are predicting rare fraud, 0.5% positive. Which metric would you optimise and why?'

Strong answer: 'Accuracy is misleading because always-predicting-negative scores 99.5%. I would look at precision-recall AUC and pick an operating point based on the business cost of missed fraud versus false alarms. If manual review is expensive, I would favour precision; if missed fraud is costly, I would raise recall and accept more review load.' This ties the metric to a decision.

Weak answer: 'I would use accuracy, or maybe F1.' There is no reasoning about imbalance or business cost, and F1 is named without justifying its beta. The interviewer learns nothing about your judgement.

Spread practice across the four strands rather than over-preparing one.

• Refresh applied statistics: testing, intervals, and common pitfalls like multiple comparisons.
• Drill probability until conditional and expectation problems feel routine.
• Prepare to defend model and metric choices with a business rationale.
• Keep coding warm with SQL plus a scripting language.

Because interviewers push back on your reasoning, rehearse defending answers aloud. A free AI voice mock interview on InterviewPrep generates a data-science mock from your CV and a target job description, then scores your reasoning delivery, pace and filler words, so you get used to explaining trade-offs calmly rather than freezing when challenged.