Master of Data Science Exam — Questions and Answers

Question 1: Which characteristic of big data refers to the uncertainty and unreliability of data sources?

Velocity
Volume
Veracity (Correct answer)
Variety

Correct answer: Veracity

Veracity addresses data quality concerns such as noise, bias, and trustworthiness inherent in large heterogeneous data sources.

Question 2: Identify the accurate statement.

The choice of an appropriate metric will influence the shape of the clusters
Hierarchical clustering is also called HCA
All of the above (Correct answer)
In general, the merges and splits are determined in a greedy manner

Correct answer: All of the above

All the statements are accurate regarding hierarchical clustering. It is indeed often referred to as Hierarchical Cluster Analysis (HCA). The merges or splits are typically determined in a greedy manner, making locally optimal decisions at each step. Furthermore, the choice of distance or similarity metric significantly influences how clusters are formed and their resulting shape.

Question 3: When using Spark Structured Streaming, what does 'output mode: complete' mean?

The stream stops after one batch is complete
The entire result table is written to the sink on every trigger (Correct answer)
Only updated rows are written
Only the rows added since the last trigger are written

Correct answer: The entire result table is written to the sink on every trigger

In complete mode, the full aggregated result table is rewritten to the output sink every time a trigger fires.

Question 4: Which of the following uses additive logistic regression as the foundation for statistical boosting?

gamBoost (Correct answer)
gbm
mboost
ada

Correct answer: gamBoost

The `gamBoost` function is specifically designed for generalized additive models (GAMs) and utilizes boosting techniques. It builds upon the principles of additive logistic regression, iteratively fitting simple models to residuals to improve overall prediction, making it the correct option among those listed for statistical boosting based on this foundation.

Question 5: A data scientist wants to show how a company's market share changed across five product categories over three years. Which visualization is most suitable?

Box plot
Scatter plot
Pie chart
Grouped bar chart (Correct answer)

Correct answer: Grouped bar chart

A grouped bar chart allows direct comparison of multiple categories across multiple time periods simultaneously, making it easy to see both within-year category differences and year-over-year trends for each category.

Question 6: You are tasked with exploring a high-dimensional dataset with over 100 features to understand its underlying structure and to simplify it for future modeling. Which EDA technique is primarily used for dimensionality reduction by transforming a large set of variables into a smaller one that still contains most of the information?

Outlier Detection using Z-scores
Cluster Analysis
Box-Cox Transformation
Principal Component Analysis (PCA) (Correct answer)

Correct answer: Principal Component Analysis (PCA)

Principal Component Analysis (PCA) is a dimensionality reduction technique used to transform a large set of correlated variables into a smaller set of uncorrelated variables called principal components, while retaining most of the original information (variance). This makes it ideal for simplifying high-dimensional data during EDA. Cluster analysis is for grouping similar data points, Box-Cox is for transforming data to a normal distribution, and Z-scores are for identifying outliers, not for dimensionality reduction.

Question 7: What is the main advantage of using a violin plot over a standard box plot?

It displays exact data point values
It calculates statistical significance
It removes all outliers automatically
It shows the full probability density of the data (Correct answer)

Correct answer: It shows the full probability density of the data

Violin plots combine a box plot with a kernel density estimate, revealing the full shape of the distribution including multimodality.

Question 8: In accessible data visualization, what is the purpose of providing an 'alt text' description alongside a chart image?

To improve SEO ranking of the page
To store the raw data in the HTML for machine parsing
To display a caption below the chart in all browsers
To convey the chart's key insight to screen-reader users who cannot see the image (Correct answer)

Correct answer: To convey the chart's key insight to screen-reader users who cannot see the image

Alt text for charts should describe the type of chart and its key finding so blind and low-vision users consuming the page via screen readers receive equivalent information.

Question 9: A machine learning model is developed to detect a rare but critical disease. For this application, failing to identify a person who has the disease (a false negative) is a far more severe error than incorrectly flagging a healthy person (a false positive). Which evaluation metric should be prioritized for optimization?

Recall (Sensitivity) (Correct answer)
Accuracy
F1-Score
Precision

Correct answer: Recall (Sensitivity)

Recall, also known as Sensitivity or True Positive Rate, measures the model's ability to find all actual positive cases (TP / (TP + FN)). In a critical medical diagnosis scenario, maximizing recall is crucial to minimize the number of dangerous false negatives.

Question 10: What distinguishes a generative model from a discriminative model?

Generative models learn the joint distribution P(X,Y); discriminative models learn P(Y|X) directly (Correct answer)
Generative models only work with continuous data; discriminative models work with categorical data
Generative models require labeled data; discriminative models do not
Generative models use gradient descent; discriminative models use closed-form solutions

Correct answer: Generative models learn the joint distribution P(X,Y); discriminative models learn P(Y|X) directly

Generative models model how data is generated (joint distribution), enabling them to synthesize new samples, while discriminative models focus solely on the decision boundary P(Y|X).

Question 11: What is the time complexity of training a k-Nearest Neighbors classifier on n training samples with d features?

O(n log n)
O(d log d)
O(nd) training, O(1) at prediction
O(1) training, O(nd) at prediction (Correct answer)

Correct answer: O(1) training, O(nd) at prediction

kNN is a lazy learner — training is O(1) since it just stores data, but each prediction requires computing distances to all n points across d features.

Question 12: Which Hadoop component is responsible for resource management and job scheduling across a cluster?

HDFS
YARN (Correct answer)
MapReduce
HBase

Correct answer: YARN

YARN (Yet Another Resource Negotiator) manages cluster resources and schedules jobs in Hadoop 2.x and later.

Question 13: Which technique addresses class imbalance at the feature level by creating synthetic minority samples?

SMOTE (Synthetic Minority Over-sampling Technique) (Correct answer)
Weight of Evidence encoding
PCA whitening
Variance Inflation Factor reduction

Correct answer: SMOTE (Synthetic Minority Over-sampling Technique)

SMOTE generates synthetic minority-class samples by interpolating between existing minority-class instances in feature space, improving class balance without simply duplicating samples.

Question 14: In Principal Component Analysis (PCA), what do the eigenvalues of the covariance matrix represent?

The mean of each transformed feature
The direction of maximum variance in the data
The amount of variance explained by each principal component (Correct answer)
The correlation between original features

Correct answer: The amount of variance explained by each principal component

Eigenvalues represent the variance explained by each principal component. The eigenvectors (not eigenvalues) represent the directions of maximum variance. This is why eigenvalues are used to decide how many components to retain.

Question 15: Which resampling technique for handling class imbalance generates synthetic minority samples by interpolating between existing minority observations?

Bootstrap aggregation
Random undersampling
SMOTE (Synthetic Minority Over-sampling Technique) (Correct answer)
Stratified k-fold cross-validation

Correct answer: SMOTE (Synthetic Minority Over-sampling Technique)

SMOTE creates new synthetic examples by interpolating between nearest neighbors in the minority class, increasing representation without simple duplication.

Question 16: A significant drawback of the K-Means clustering algorithm is its sensitivity to the initial placement of cluster centroids. A poor initialization can lead to a suboptimal clustering solution. Which of the following is a common method used to mitigate this issue?

Scaling the data using Min-Max normalization before clustering.
Using a dendrogram to visualize the clusters.
Running the algorithm multiple times with different random initializations and choosing the best result. (Correct answer)
Applying the elbow method to determine the optimal K.

Correct answer: Running the algorithm multiple times with different random initializations and choosing the best result.

Because K-Means can converge to a local optimum depending on the random starting positions of its centroids, a common and effective strategy is to run the algorithm multiple times (n_init) with different random initializations. The final result is then chosen from the run that yields the best performance, typically measured by the lowest within-cluster sum of squares (inertia). The elbow method helps find K but doesn't solve the initialization problem. Scaling is good practice but doesn't directly address initialization sensitivity. Dendrograms are associated with hierarchical clustering.

Question 17: What is the Calinski-Harabasz index, and what does a higher value indicate?

It computes the log-likelihood of a GMM; higher values indicate more components are needed
It is the ratio of between-cluster dispersion to within-cluster dispersion; higher values indicate better-defined clusters (Correct answer)
It measures cluster overlap; higher values indicate more overlapping clusters
It measures the average silhouette width; higher values indicate more noise points

Correct answer: It is the ratio of between-cluster dispersion to within-cluster dispersion; higher values indicate better-defined clusters

The Calinski-Harabasz index (also called Variance Ratio Criterion) rewards tight, well-separated clusters: higher values mean clusters are dense internally and far from each other.

Question 18: Data that list all findings in a category are referred to as _________ data.

summarized (Correct answer)
None of the above
raw
frequency

Correct answer: summarized

Summarized data presents aggregated information, often in tables or charts, that condenses raw findings into meaningful categories. This process involves calculating totals, averages, or counts to provide an concise overview rather than individual data points, making it easier to understand trends and patterns.

Question 19: A dataset containing sensor readings has missing values in a 'Temperature' column. The data is missing completely at random (MCAR), and the overall temperature distribution is not heavily skewed. To handle the missing data, a data scientist decides to use an imputation method. Which of the following is a simple, yet reasonable, imputation strategy in this scenario?

Replace missing values with the mean of the 'Temperature' column (Correct answer)
Use K-Nearest Neighbors (KNN) Imputation
Delete all rows with missing temperature values
Replace missing values with a constant like 0

Correct answer: Replace missing values with the mean of the 'Temperature' column

For numerical data that is missing completely at random and has a symmetric distribution, mean imputation is a straightforward and common technique. [17, 22] It replaces the missing values with the average of the observed values, which can be a reasonable estimate that doesn't significantly distort the overall distribution. [18] Deleting rows can lead to a significant loss of information, and replacing with a constant can introduce bias. KNN Imputation is a more complex method that, while powerful, might be unnecessarily complex for this simple case.

Question 20: Identify the replication-related correct statement.

Focuses on the validity of the scientific claim
Arguably a minimum standard for any scientific study
All of the above
Focuses on the validity of the data analysis (Correct answer)

Correct answer: Focuses on the validity of the data analysis

Replication in scientific studies involves repeating the *analysis* of a study, often with new or different data, to see if the original findings hold. Its primary focus is to validate the methods, computational steps, and statistical models used in the original data analysis. This ensures the robustness and correctness of the analytical process, rather than directly validating the broader scientific claim itself, which is more the domain of reproducibility and independent studies.

Question 21: In the context of big data, what does the term 'data lake' primarily refer to?

A cloud-based data warehouse with predefined schemas
A centralized repository storing raw data in its native format (Correct answer)
A relational database optimized for analytics
A real-time streaming data pipeline

Correct answer: A centralized repository storing raw data in its native format

A data lake stores vast amounts of raw data in its original format until it is needed for analysis.

Question 22: A data scientist splits a dataset into 80% training and 20% test sets, trains a complex neural network, and reports excellent test accuracy. A colleague suggests the model may still be overfit. Which additional evaluation strategy would best reveal this?

Use k-fold cross-validation on the full dataset and compare variance across folds (Correct answer)
Report precision and recall instead of accuracy
Apply dimensionality reduction before retraining
Increase the test set size to 40% and retrain

Correct answer: Use k-fold cross-validation on the full dataset and compare variance across folds

K-fold cross-validation assesses model stability across multiple train/test splits. High variance in scores across folds indicates overfitting that a single train/test split can miss, since that split may have been favorable by chance.

Question 23: In maximum likelihood estimation, the score function is defined as:

The expected value of the log-likelihood
The second derivative of the log-likelihood
The gradient of the log-likelihood with respect to the parameter (Correct answer)
The ratio of the likelihood to the prior

Correct answer: The gradient of the log-likelihood with respect to the parameter

The score is ∂ℓ(θ)/∂θ, the first derivative of the log-likelihood; at the MLE it equals zero.

Question 24: What is the primary purpose of Maximum Likelihood Estimation (MLE)?

To find the parameter values that minimize the squared residuals between observations and predictions
To find the parameter values that maximize the probability of observing the given data under the assumed model (Correct answer)
To compute the posterior distribution of parameters given prior beliefs and data
To find parameter values that minimize the Kullback-Leibler divergence from a reference distribution

Correct answer: To find the parameter values that maximize the probability of observing the given data under the assumed model

MLE selects the parameter values θ that maximize the likelihood function L(θ | data) = P(data | θ), making the observed data most probable under the model. Minimizing squared residuals is OLS (equivalent to MLE only under Gaussian errors), and maximizing the posterior is MAP estimation.

Question 25: Which algorithm is commonly used for collaborative filtering in recommendation systems?

K-means clustering
Matrix factorization (e.g., SVD) (Correct answer)
DBSCAN
Naive Bayes

Correct answer: Matrix factorization (e.g., SVD)

Matrix factorization decomposes the user-item interaction matrix into latent factor matrices, capturing hidden preference patterns.

Question 26: During model validation, what problem does data leakage most commonly cause?

Underfitting due to reduced training set size
Overly optimistic performance estimates that fail to replicate on new data (Correct answer)
Increased computational cost during cross-validation
Lower precision but higher recall on the test set

Correct answer: Overly optimistic performance estimates that fail to replicate on new data

Data leakage allows information from the test set to influence training, resulting in inflated performance metrics that do not generalize.

Question 27: Which technique is used to assess the stability of a regression model by partitioning data into training and validation subsets multiple times?

Stratified random sampling
Bootstrap sampling
Jackknife resampling
K-fold cross-validation (Correct answer)

Correct answer: K-fold cross-validation

K-fold cross-validation splits data into k subsets and iteratively trains/validates across all folds to estimate model performance.

Question 28: Which technique helps address class imbalance in a binary classification dataset?

SMOTE (Synthetic Minority Over-sampling Technique) (Correct answer)
Adding more hidden layers
Using a larger batch size
Increasing the learning rate

Correct answer: SMOTE (Synthetic Minority Over-sampling Technique)

SMOTE generates synthetic examples of the minority class by interpolating between existing minority samples, helping the model learn the underrepresented class better.

Question 29: Which of the following uses information about one object to forecast the values of another?

Exploratory
Inferential
Predictive (Correct answer)
None of the above

Correct answer: Predictive

Predictive analysis focuses on using historical data, statistical algorithms, and machine learning techniques to forecast future outcomes or unknown values. It involves building models that can predict a target variable based on the relationships identified with other input variables. This is distinct from exploratory, inferential, or descriptive analysis.

Question 30: In a random forest, what is the purpose of feature randomness (selecting a subset of features at each split)?

It ensures all features contribute equally to every tree
It decorrelates the individual trees, reducing ensemble variance (Correct answer)
It prevents trees from exceeding their maximum depth
It speeds up training by reducing the number of computations per split

Correct answer: It decorrelates the individual trees, reducing ensemble variance

By considering only a random subset of features at each split, trees in a random forest are de-correlated, so their errors are less likely to coincide, reducing overall variance.

Question 31: In byte-pair encoding (BPE) tokenization, what determines when to stop merging character pairs?

When a predefined vocabulary size limit is reached (Correct answer)
When the vocabulary contains no more than 256 entries
When all tokens are single Unicode characters
When token frequency drops below 1

Correct answer: When a predefined vocabulary size limit is reached

BPE iteratively merges the most frequent pair of adjacent symbols until the target vocabulary size is achieved.

Question 32: Which technique specifically tests whether a model has learned spurious correlations by evaluating it on data where the target variable is shuffled?

Stratified sampling
Bootstrap validation
Ablation study
Permutation test (Correct answer)

Correct answer: Permutation test

A permutation test shuffles the labels and measures model performance; if the model scores high on shuffled data, it likely learned noise rather than true signal.

Question 33: What is the primary purpose of the Expectation-Maximization (EM) algorithm?

To compute Bayesian credible intervals for regression coefficients
To test for stationarity in time series data
To perform dimensionality reduction on high-dimensional data
To find maximum likelihood estimates when data has missing or latent variables (Correct answer)

Correct answer: To find maximum likelihood estimates when data has missing or latent variables

The EM algorithm iterates between computing expected values of latent variables (E-step) and maximizing the likelihood given those expectations (M-step) to find MLEs with incomplete data.

Question 34: In the context of AdaBoost, how are training samples reweighted across iterations?

Correctly classified samples get higher weights to reinforce learning
All samples are reweighted uniformly to maintain equal contribution
Misclassified samples receive higher weights so subsequent classifiers focus on hard examples (Correct answer)
Weights are assigned based on feature importance scores from previous rounds

Correct answer: Misclassified samples receive higher weights so subsequent classifiers focus on hard examples

AdaBoost increases the weights of misclassified samples after each round, forcing the next weak learner to concentrate on the examples the current ensemble handles poorly.

Question 35: What does a Box-Cox transformation accomplish in feature engineering?

It applies a piecewise linear transformation to cap outliers at specified percentiles
It centers and scales features to zero mean and unit variance
It converts categorical variables into continuous probability scores
It maps a positive-valued feature to a more normal distribution by finding an optimal power parameter λ (Correct answer)

Correct answer: It maps a positive-valued feature to a more normal distribution by finding an optimal power parameter λ

Box-Cox finds the optimal power λ that makes a positive feature's distribution as close to normal as possible, stabilizing variance and improving model assumptions.

Question 36: When conducting a literature review for a data science thesis, which database is most specialized for computer science and data science research papers?

IEEE Xplore and ACM Digital Library (Correct answer)
LexisNexis
JSTOR
PubMed

Correct answer: IEEE Xplore and ACM Digital Library

IEEE Xplore and the ACM Digital Library are the primary repositories for peer-reviewed computer science, machine learning, and data science research.

Question 37: Identify the accurate statement.

None of the above
NULL is the standard missing data marker used in S
Frequency inference is the use of Bayesian probability representation of beliefs to perform inference
Bayesian inference is the use of Bayesian probability representation of beliefs to perform inference (Correct answer)

Correct answer: Bayesian inference is the use of Bayesian probability representation of beliefs to perform inference

Bayesian inference is a statistical method that uses Bayes' theorem to update the probability for a hypothesis as more evidence or information becomes available. It fundamentally relies on the Bayesian interpretation of probability, where probabilities represent degrees of belief. This allows for a coherent framework to reason about uncertainty and make decisions based on updated beliefs.

Question 38: What is the 'elbow method' limitation when used to select k for k-means clustering?

It requires rerunning k-means for every possible value of k from 1 to n
The elbow is often ambiguous or absent, making objective k selection difficult in practice (Correct answer)
It only works when clusters have equal variance and size
It maximizes the silhouette score rather than minimizing inertia

Correct answer: The elbow is often ambiguous or absent, making objective k selection difficult in practice

Real-world inertia curves often decrease smoothly without a clear inflection point, making the elbow subjective and unreliable for automated k selection.

Question 39: Centered at ___, normalized data have units equal to the standard deviations of the original data.

5
1
10
0 (Correct answer)

Correct answer: 0

Normalized data, typically achieved through standardization (e.g., z-score transformation), involves subtracting the mean and dividing by the standard deviation. This process results in data with a mean of 0 and a standard deviation of 1. Therefore, normalized data are always centered at 0.

Question 40: In maximum likelihood estimation, what is the Fisher information used to approximate?

The bias of the estimator
The p-value of a hypothesis test
The posterior probability of the parameter
The variance of the MLE asymptotically (Correct answer)

Correct answer: The variance of the MLE asymptotically

The inverse of the Fisher information provides the asymptotic variance of the MLE, forming the basis of the Cramér-Rao lower bound.

Question 41: Which of the following languages is crucial for data science?

R (Correct answer)
None of the above
Java
Ruby

Correct answer: R

R is a powerful and widely used programming language specifically designed for statistical computing and graphics. It offers an extensive ecosystem of packages for data manipulation, visualization, statistical modeling, and machine learning, making it a crucial tool for many data scientists. While Python is also very popular, R remains a cornerstone in the field.

Question 42: What is the purpose of annotation layers in exploratory data visualizations?

To replace axis labels with more descriptive text
To highlight key data points, trends, or anomalies with contextual notes (Correct answer)
To add decorative elements that improve aesthetics
To embed raw data values into every data point

Correct answer: To highlight key data points, trends, or anomalies with contextual notes

Annotations direct the viewer's attention to significant patterns or outliers and provide context that the data alone may not convey.

Question 43: When visualizing geospatial data showing values per region, which map projection concern is most critical?

Color rendering differences across browsers
Area distortion that can misrepresent the magnitude of values in different regions (Correct answer)
Whether the map includes a compass rose
The number of labels that can fit on the map

Correct answer: Area distortion that can misrepresent the magnitude of values in different regions

Choropleth maps using projections with significant area distortion can make smaller regions appear less important regardless of their actual data values.

Question 44: Which of the following is a primary advantage of Hierarchical Clustering over K-Means Clustering?

It does not require the number of clusters to be specified beforehand. (Correct answer)
It always converges to the same solution regardless of initialization.
It can handle categorical features without any transformation.
It is more computationally efficient for large datasets.

Correct answer: It does not require the number of clusters to be specified beforehand.

A key advantage of Hierarchical Clustering is that it does not require the user to pre-specify the number of clusters. It produces a dendrogram, a tree-like diagram that visualizes the cluster hierarchy, allowing the user to choose the number of clusters after the model is built. K-Means, in contrast, requires the number of clusters (K) as an input parameter. Hierarchical clustering is generally more computationally expensive than K-Means for large datasets, and K-Means is sensitive to initial centroid placement.

Question 45: In Apache Spark, what is the 'Catalyst optimizer' responsible for?

Managing memory allocation for RDD caching
Transforming logical query plans into optimized physical execution plans for DataFrames and Datasets (Correct answer)
Compressing shuffle data between executors
Scheduling tasks across YARN worker nodes

Correct answer: Transforming logical query plans into optimized physical execution plans for DataFrames and Datasets

Catalyst is Spark SQL's rule-based and cost-based query optimizer that rewrites logical plans into efficient physical plans.

Question 46: When using bootstrap validation (0.632+ bootstrap), what does the 0.632 factor represent?

The confidence level of the performance estimate
The expected proportion of unique training samples in each bootstrap sample (Correct answer)
The minimum required accuracy for model acceptance
The ratio of test samples to training samples

Correct answer: The expected proportion of unique training samples in each bootstrap sample

In a bootstrap sample of size n drawn with replacement, approximately 63.2% of the original observations appear at least once, based on the probability 1 − (1 − 1/n)^n.

Question 47: What is the primary role of Apache ZooKeeper in a big data cluster?

Data serialization
Distributed coordination and configuration management (Correct answer)
ETL pipeline orchestration
In-memory caching

Correct answer: Distributed coordination and configuration management

Apache ZooKeeper provides distributed coordination services such as configuration management, naming, synchronization, and group membership, allowing distributed applications like Hadoop and Kafka to maintain consistency across nodes.

Question 48: A data scientist calculates a 95% confidence interval for the average user session duration to be [150, 170] seconds. Which of the following is the correct interpretation of this interval?

95% of all individual user session durations fall between 150 and 170 seconds.
The sample mean has a 95% chance of being the true population mean.
There is a 95% probability that the true population mean session duration is between 150 and 170 seconds.
If we were to repeat this sampling process many times, we would expect 95% of the calculated confidence intervals to contain the true population mean session duration. (Correct answer)

Correct answer: If we were to repeat this sampling process many times, we would expect 95% of the calculated confidence intervals to contain the true population mean session duration.

A 95% confidence interval relates to the reliability of the estimation procedure, not a specific interval. It means that if the same sampling method were used to generate many different samples and a confidence interval were calculated for each, about 95% of these intervals would capture the true, unknown population parameter. Common misinterpretations include stating there's a 95% probability the true mean is in the specific interval or that 95% of the data points lie within the interval.

Question 49: A master's student is building a classification model and notices the AUC-ROC is 0.95 on training data but 0.61 on the test set. What does this most likely indicate?

The features are not predictive
The test set has label errors
The model is overfitting to the training data (Correct answer)
The model is underfitting the data

Correct answer: The model is overfitting to the training data

A large gap between training and test performance is a classic sign of overfitting, where the model memorizes training data patterns that do not generalize.

Question 50: A data scientist wants to determine whether a new recommendation algorithm increases user engagement compared to the existing one. Which experimental design is most appropriate?

Observational cohort study
Time-series analysis without a control group
A/B testing with random user assignment (Correct answer)
Case-control study

Correct answer: A/B testing with random user assignment

A/B testing with random assignment allows causal inference by controlling for confounding variables between the treatment and control groups.

Question 51: In the context of EDA, what is the primary purpose of a quantile-quantile (Q-Q) plot?

Comparing two variables' correlation
Visualizing time series seasonality
Assessing whether data follows a theoretical distribution (Correct answer)
Identifying duplicate records

Correct answer: Assessing whether data follows a theoretical distribution

A Q-Q plot plots sample quantiles against theoretical quantiles; points falling on the diagonal line indicate the data matches that distribution.

Question 52: K-means clustering requires which of the following?

initial guess as to cluster centroids
All of the above (Correct answer)
defined distance metric
number of clusters

Correct answer: All of the above

K-means clustering requires several inputs to function. Users must specify the desired `number of clusters` (k) beforehand. It also needs an `initial guess as to cluster centroids` to begin its iterative process. Finally, a `defined distance metric` (typically Euclidean distance) is essential to measure similarity for point assignment and centroid recalculation.

Question 53: Which of the following is a core limitation of PCA for dimensionality reduction in the context of unsupervised learning?

PCA cannot handle datasets with more features than samples
PCA only captures linear relationships and misses nonlinear structure in the data (Correct answer)
PCA requires specifying the number of components before seeing the data
PCA is computationally infeasible for datasets with fewer than 1,000 dimensions

Correct answer: PCA only captures linear relationships and misses nonlinear structure in the data

PCA finds orthogonal directions of maximum variance using linear projections, so it cannot capture curved manifolds or other nonlinear structure in high-dimensional data.

Question 54: Which of the following data wrangling tasks involves creating new variables from existing ones to better represent the underlying patterns in the data for a machine learning model?

Data Integration
Data Validation
Data Cleaning
Feature Engineering (Correct answer)

Correct answer: Feature Engineering

Feature engineering is the process of using domain knowledge to create new features (variables) from the raw data that make machine learning algorithms work better. Data cleaning deals with errors and missing values, integration combines data sources, and validation checks data quality.

Question 55: Which metric best evaluates whether a data visualization successfully communicates its intended message?

The visual complexity score of the design
The number of chart types used in the visualization
The total amount of data points displayed
Task completion accuracy when users answer questions about the visualized data (Correct answer)

Correct answer: Task completion accuracy when users answer questions about the visualized data

Measuring whether users correctly interpret the data and answer questions validates that the visualization achieves its communication goal.

Question 56: In the transformer architecture, what does the term 'positional encoding' accomplish?

It masks future tokens during training
It injects information about each token's position in the sequence (Correct answer)
It normalizes token embeddings to unit length
It reduces the dimensionality of the key-query product

Correct answer: It injects information about each token's position in the sequence

Because transformers process all tokens in parallel without recurrence, positional encodings (sinusoidal or learned vectors) are added to embeddings so the model can distinguish token order.

Question 57: Which algorithm is commonly used to segment text into subword units, helping models handle out-of-vocabulary words?

Viterbi decoding
Latent Dirichlet Allocation
Byte Pair Encoding (BPE) (Correct answer)
TF-IDF weighting

Correct answer: Byte Pair Encoding (BPE)

Byte Pair Encoding iteratively merges the most frequent character pairs to build a subword vocabulary, allowing models to represent rare or unseen words as combinations of known subword tokens.

Question 58: The Brier score is used to evaluate which type of model output?

Feature importance rankings
Regression residuals
Probability estimates from a probabilistic classifier (Correct answer)
Class labels from a classifier

Correct answer: Probability estimates from a probabilistic classifier

The Brier score measures the mean squared difference between predicted probabilities and actual binary outcomes, assessing probabilistic calibration.

Question 59: What is the purpose of A/B testing in a data-driven organization?

To visualize data distributions
To clean and preprocess raw datasets
To compare two versions of a variable to determine which performs better (Correct answer)
To train machine learning models on labeled data

Correct answer: To compare two versions of a variable to determine which performs better

A/B testing is a controlled experiment where two variants are compared to measure which one produces a better outcome on a specific metric.

Question 60: Which of the following is the correct interpretation of a 95% confidence interval for model accuracy?

95% of test samples were correctly classified
There is a 95% probability the true accuracy lies in this interval
The model will be accurate 95% of the time on new data
If the experiment were repeated many times, 95% of such intervals would contain the true accuracy (Correct answer)

Correct answer: If the experiment were repeated many times, 95% of such intervals would contain the true accuracy

A confidence interval is a frequentist concept meaning that the procedure generates intervals containing the true parameter 95% of the time across repeated experiments.

Question 61: Effect size measures like Cohen's d are reported alongside p-values because:

They are required by all journals
They replace the need for significance testing
They quantify the practical magnitude of a difference, which p-values do not (Correct answer)
They eliminate sampling error

Correct answer: They quantify the practical magnitude of a difference, which p-values do not

A statistically significant result can have a trivially small effect; Cohen's d communicates how large the difference actually is in standardized units.

Master of Data Science Exam

The Master of Data Science Exam exam validates essential knowledge and skills required for certification or licensure in this field.

Questions

60 min

Time Limit

70%

To Pass

Checking for saved progress...

Exam Rules

You can skip questions and return to them later
Flag questions for review before submitting
No feedback shown until you submit the entire exam
Unanswered questions count as wrong — answer everything
10 pretest questions are mixed in and don't affect your score
Timer auto-submits when time runs out
Your progress is auto-saved every 30 seconds