Data Scientist Interview Questions & Answers

Sample Answer

Bias is the error from incorrect assumptions in the learning algorithm — high bias leads to underfitting. Variance is the error from sensitivity to fluctuations in the training set — high variance leads to overfitting. The trade-off: as model complexity increases, bias decreases but variance increases, and vice versa. The goal is to find the sweet spot that minimises total error. Regularisation techniques (L1/L2) and cross-validation help manage this trade-off in practice.

Sample Answer

Supervised learning trains on labelled data — you have inputs and corresponding correct outputs (labels). The model learns to map inputs to outputs. Examples: classification, regression. Unsupervised learning finds patterns in unlabelled data — no correct outputs are provided. Examples: clustering (K-means), dimensionality reduction (PCA), anomaly detection. Semi-supervised learning combines both: a small amount of labelled data with a large amount of unlabelled data.

Sample Answer

Several techniques depending on severity: (1) Resampling — oversample the minority class (SMOTE) or undersample the majority class. (2) Cost-sensitive learning — assign higher misclassification costs to the minority class. (3) Use appropriate evaluation metrics — accuracy is misleading with imbalanced data; use F1-score, precision-recall AUC, or ROC AUC instead. (4) Use algorithms that handle imbalance well (e.g., tree-based methods, SVMs). The best approach depends on the business cost of false positives vs false negatives.

Sample Answer

A p-value is the probability of observing results as extreme as those in your data, assuming the null hypothesis is true. A low p-value (typically < 0.05) suggests the result is unlikely due to chance. Limitations: it doesn't measure effect size or practical significance; it's binary (significant/not) when reality is continuous; it's susceptible to p-hacking with multiple comparisons; and it requires a pre-specified hypothesis — post-hoc analysis inflates false positive rates.

Sample Answer

First, understand the business objective — what decision will this analysis inform? Then define the problem in data terms: what is the target variable, what data is available, and what are the constraints? Explore and clean the data (EDA). Choose appropriate methods. Model, evaluate with the right metrics, and iterate. Finally, communicate findings in business terms — not model accuracy, but expected impact on revenue, churn, or whatever the business cares about.

Sample Answer

Pick a real project with a clear business impact. Structure your answer: the business problem, your approach (data sources, methods chosen and why), key challenges you overcame, and the result in business terms. Bonus points for mentioning what you'd do differently in hindsight — it shows maturity and self-reflection.