Question 1

Explain the difference between Type I and Type II errors.

Accepted Answer

Type I Error (False Positive, α): Rejecting a true null hypothesis. Example: A medical test says you have a disease when you don't. Type II Error (False Negative, β): Failing to reject a false null hypothesis. Example: A test says you're healthy when you actually have the disease. Trade-off: Reducing one increases the other. In medical testing, minimizing Type II (missing a disease) is critical. In spam filtering, minimizing Type I (real email marked as spam) is important. Significance level α (usually 0.05) controls Type I error rate. Power (1-β) is the probability of correctly rejecting a false null hypothesis.

Question 2

What is the Central Limit Theorem and why is it important?

Accepted Answer

The CLT states that the sampling distribution of the mean approaches a normal distribution as sample size increases, regardless of the population's original distribution (with finite mean and variance). Practically important because: 1) Allows use of z-tests and confidence intervals even for non-normal data (n≥30 is general guideline); 2) Foundation for statistical inference; 3) Explains why many natural phenomena appear normally distributed. Example: Average heights from many random samples of 50 people will form a bell curve, even if the individual height distribution is slightly skewed.

Question 3

Explain p-value in simple terms.

Accepted Answer

The p-value is the probability of observing results as extreme as (or more extreme than) the actual results, assuming the null hypothesis is true. If p < 0.05 (common threshold): The observed result is unlikely under the null hypothesis → reject null. If p > 0.05: Insufficient evidence to reject null. Important nuances: p-value is NOT the probability that the null hypothesis is true. A small p-value doesn't mean the effect is large or practical. With very large samples, even tiny meaningless effects can be "statistically significant." Always combine p-values with effect sizes and confidence intervals.

Question 4

Explain the bias-variance tradeoff.

Accepted Answer

Bias: Error from overly simplistic model assumptions (underfitting). High bias → model misses patterns. Example: Linear regression on non-linear data. Variance: Error from sensitivity to small fluctuations in training data (overfitting). High variance → model memorizes noise. Example: Deep decision tree on small dataset. Total Error = Bias² + Variance + Irreducible Error. The goal is finding the sweet spot: complex enough to capture patterns (low bias) but general enough to work on new data (low variance). Techniques: Regularization (L1/L2), cross-validation, ensemble methods (Random Forest reduces variance, Boosting reduces bias), more training data.

Question 5

What is the difference between supervised and unsupervised learning?

Accepted Answer

Supervised Learning: Training data has labeled inputs and outputs. Model learns input→output mapping. Types: Classification (predict category — spam/not spam) and Regression (predict continuous value — house price). Algorithms: Linear Regression, Decision Trees, SVM, Neural Networks. Unsupervised Learning: No labels. Model finds hidden patterns/structure. Types: Clustering (group similar data — K-Means, DBSCAN), Dimensionality Reduction (PCA, t-SNE), Association Rules. Also: Semi-supervised (some labels), Self-supervised (create labels from data), Reinforcement Learning (learn from rewards/penalties). Real-world: Most production ML is supervised; unsupervised is used for EDA and feature engineering.

Question 6

Explain regularization (L1 vs L2).

Accepted Answer

Regularization adds a penalty to the loss function to prevent overfitting. L1 (Lasso): Adds absolute value of coefficients (|w|) as penalty. Produces sparse models — many coefficients become exactly zero. Great for feature selection. L2 (Ridge): Adds squared coefficients (w²) as penalty. Shrinks all coefficients toward zero but doesn't eliminate them. Better when all features contribute. Elastic Net: Combines L1 + L2 penalties. Hyperparameter λ controls regularization strength: λ=0 → no regularization (overfit risk), λ→∞ → underfitting. Use cross-validation to find optimal λ.

Question 7

What is cross-validation and why is it used?

Accepted Answer

Cross-validation estimates how well a model generalizes to unseen data. K-Fold CV: Split data into K parts. Train on K-1 folds, test on remaining fold. Repeat K times. Average the K performance metrics. Common: K=5 or K=10. Stratified K-Fold: Preserves class distribution in each fold — essential for imbalanced data. Leave-One-Out (LOO): K = n. Expensive but useful for small datasets. Why it matters: 1) More reliable performance estimate than single train-test split; 2) Uses all data for both training and validation; 3) Helps detect overfitting; 4) Used in hyperparameter tuning (Grid/Random Search with CV).

Question 8

What is the difference between Pandas DataFrame and NumPy Array?

Accepted Answer

NumPy Array: Homogeneous data type (all elements same type), supports fast vectorized mathematical operations, N-dimensional, indexing by integer position. Best for: Numerical computation, linear algebra, matrix operations. Pandas DataFrame: Heterogeneous (columns can have different types), labeled axes (row & column names), built on top of NumPy, supports SQL-like operations (join, groupby, pivot). Best for: Data manipulation, cleaning, exploration, time series. Performance: NumPy is faster for pure numerical computation. Pandas adds overhead for labels but is much more convenient for real-world data. Use NumPy arrays inside ML models, Pandas for data preparation.

Question 9

Write a SQL query to find the second highest salary from an Employee table.

Accepted Answer

Method 1 (Subquery): SELECT MAX(salary) FROM employees WHERE salary < (SELECT MAX(salary) FROM employees); Method 2 (DENSE_RANK): SELECT salary FROM (SELECT salary, DENSE_RANK() OVER (ORDER BY salary DESC) as rank FROM employees) ranked WHERE rank = 2; Method 3 (LIMIT/OFFSET): SELECT DISTINCT salary FROM employees ORDER BY salary DESC LIMIT 1 OFFSET 1; DENSE_RANK is most versatile — works for finding Nth highest and handles ties correctly. LIMIT/OFFSET is simplest but database-specific. The subquery approach is universally supported.

Data Science & Analytics Interview Questions & Answers

Statistics & Probability

Machine Learning

Python & SQL for Data Science

Key Tips for Data Science Interviews