I85PREP
⌘ K
AWS Machine Learning Specialty study materials
AWS Machine Learning Specialty study resources

AWS Machine Learning Specialty (AWS) practice questions

One question at a time. Build your AWS Machine Learning Specialty understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Data Engineering

A data engineer needs a fully managed, petabyte-scale data warehouse on AWS that supports columnar storage and massively parallel query execution. Which service should be chosen?

  1. Amazon Neptune
  2. Amazon ElastiCache
  3. Amazon Redshift Correct answer.
  4. Amazon DynamoDB

Explanation

Amazon Redshift is a managed, petabyte-scale data warehouse built on columnar storage and a massively parallel processing architecture, which makes it suited to analytic SQL workloads. Services such as DynamoDB and Neptune target operational key-value and graph access patterns rather than large-scale analytical warehousing.

Question 2Data Engineering

Which AWS service is designed as a fully managed extract, transform, and load (ETL) service that can run Spark-based jobs without managing servers?

  1. Amazon Athena
  2. AWS Glue Correct answer.
  3. AWS Data Pipeline
  4. Amazon EMR

Explanation

AWS Glue is a serverless data integration service that provisions and manages the Spark environment, offers a Data Catalog, and runs ETL jobs without cluster administration. EMR provides Spark too, but the customer manages the cluster lifecycle rather than relying on a fully serverless ETL service.

Question 3Data Engineering

A company wants to query data stored in Amazon S3 directly using standard SQL and pay only for the data scanned per query. Which service fits this requirement?

  1. Amazon Redshift Spectrum
  2. Amazon Athena Correct answer.
  3. AWS Glue DataBrew
  4. Amazon RDS

Explanation

Amazon Athena is a serverless interactive query service that runs SQL over data in S3 and charges based on the amount of data scanned. Redshift Spectrum also queries S3, but it requires an existing Redshift cluster and is not a standalone pay-per-scan query service.

Question 4Data Engineering

Which storage class in Amazon S3 is intended for long-term archival of data that can tolerate retrieval times of minutes to hours and offers the lowest storage cost?

  1. S3 Standard
  2. S3 Glacier Deep Archive Correct answer.
  3. S3 Intelligent-Tiering
  4. S3 One Zone-IA

Explanation

S3 Glacier Deep Archive is the lowest-cost storage class, designed for rarely accessed archival data with retrieval times measured in hours. Standard, Intelligent-Tiering, and One Zone-IA are all intended for data needing faster access.

Question 5Data Engineering

In Amazon Kinesis Data Streams, what is the unit of throughput capacity that a producer writes to and that determines the stream's ingestion limit?

  1. A consumer group
  2. A shard Correct answer.
  3. A partition key
  4. A delivery stream

Explanation

A Kinesis data stream is composed of shards, and each shard provides a fixed write and read capacity, so shard count determines total stream throughput. Partition keys route records to shards but are not capacity units, and delivery streams belong to Firehose rather than Data Streams.

Question 6Data Engineering

A team must load streaming click events into Amazon S3 in near real time with automatic batching, compression, and optional transformation, without writing consumer code. Which service is most appropriate?

  1. Amazon Kinesis Data Firehose Correct answer.
  2. Amazon MSK
  3. Amazon SQS
  4. Amazon Kinesis Data Streams

Explanation

Kinesis Data Firehose is a fully managed delivery service that buffers, compresses, and optionally transforms streaming records before writing them to destinations such as S3. Data Streams and MSK require the customer to build and run the consuming application, and SQS is a message queue rather than a delivery pipeline to S3.

Question 7Data Engineering

Which AWS service provides a centralized metadata repository that stores table definitions, schemas, and partition information for data in the AWS analytics ecosystem?

  1. Amazon Athena
  2. AWS Glue Data Catalog Correct answer.
  3. AWS Lake Formation
  4. AWS Schema Conversion Tool

Explanation

The AWS Glue Data Catalog is a persistent metadata store holding database and table definitions, schemas, and partition metadata that services such as Athena, Redshift Spectrum, and EMR can share. Lake Formation governs permissions over that catalog but is not itself the metadata repository.

Question 8Data Engineering

A data pipeline must move data from an on-premises Oracle database to Amazon S3 on a schedule with minimal operational overhead. Which AWS service is purpose-built for this managed data movement?

  1. AWS Glue DataBrew
  2. AWS Database Migration Service Correct answer.
  3. Amazon EventBridge
  4. Amazon Kinesis Data Firehose

Explanation

AWS Database Migration Service can replicate or migrate data from on-premises relational databases to targets such as S3, and it is managed so operational overhead is low. Firehose handles streaming ingestion, DataBrew is for data preparation, and EventBridge only routes events.

Question 9Data Engineering

When designing an Amazon Kinesis Data Streams producer, which partition key strategy best distributes records evenly across shards?

  1. Use a high-cardinality key such as a random UUID or session identifier Correct answer.
  2. Use a constant key so all records land in one shard
  3. Use the shard ID assigned by the stream
  4. Use the record timestamp truncated to the day

Explanation

Kinesis hashes the partition key to choose a shard, so a high-cardinality key spreads records uniformly and avoids hot shards. A constant or low-cardinality key concentrates traffic on a single shard and throttles throughput.

Question 10Data Engineering

An Athena query over a large S3 dataset is slow and expensive. The data is stored as uncompressed CSV and is not partitioned. Which combination of changes most directly reduces both query latency and bytes scanned?

  1. Increase the Athena query result reuse window and add more S3 buckets
  2. Convert to a columnar format such as Parquet and add partition columns in the table definition Correct answer.
  3. Increase the number of workgroups and set a higher per-query data limit
  4. Enable S3 versioning and move objects to Glacier Deep Archive

Explanation

Athena bills by data scanned, so columnar formats like Parquet allow column pruning and compression, while partitioning enables partition pruning that skips irrelevant data. Together these reduce both bytes scanned and query runtime.

Question 11Data Engineering

A Redshift cluster runs a nightly ETL load followed by heavy reporting queries. The team wants to reduce the time spent vacuuming and improve sort performance. Which design choice best supports this?

  1. Remove all sort keys and rely on the default distribution style
  2. Convert all tables to temporary tables before loading
  3. Disable automatic analyze so statistics stay static
  4. Define sort keys on frequently filtered columns and let automatic vacuum and analyze run Correct answer.

Explanation

Choosing sort keys that match common filter columns lets Redshift skip blocks during scans, and automatic vacuum and analyze maintain physical ordering and statistics with less manual effort. Removing keys or freezing statistics degrades the query planner's ability to prune data.

Question 12Data Engineering

A company must give an analytics team read access to only specific columns of a table registered in the AWS Glue Data Catalog, while other teams see all columns. Which service enforces this column-level access control?

  1. AWS Identity and Access Management only
  2. AWS Lake Formation Correct answer.
  3. AWS Glue crawler classifications
  4. Amazon S3 bucket policies

Explanation

Lake Formation provides fine-grained permissions, including column-level and row-level filters, over Data Catalog tables. IAM and S3 bucket policies operate at the resource or object level and cannot express column-level restrictions on catalog tables.

Question 13Data Engineering

A streaming application reads from a Kinesis data stream with four shards. What is the approximate maximum total read throughput available to consumers using the shared fan-out model?

  1. 20 MB per second
  2. 2 MB per second
  3. 8 MB per second Correct answer.
  4. 4 MB per second

Explanation

Each Kinesis shard supports up to 2 MB per second of read throughput in the shared throughput model, so four shards provide roughly 8 MB per second in total. Enhanced fan-out provides dedicated throughput per consumer but changes the model.

Question 14Data Engineering

A data engineer needs to run a Spark job that processes 50 TB nightly and requires fine control over cluster sizing, instance types, and installed libraries. Which service offers the most control while remaining managed?

  1. AWS Glue with serverless Spark
  2. Amazon Athena
  3. Amazon Kinesis Data Analytics
  4. Amazon EMR Correct answer.

Explanation

Amazon EMR lets teams choose instance types, cluster size, and bootstrap actions to install libraries, giving fine control over the Spark environment. Glue serverless Spark abstracts away cluster sizing, and Athena and Kinesis Data Analytics are not general-purpose Spark cluster platforms.

Question 15Data Engineering

An S3-based data lake stores small JSON files arriving every minute. Downstream Athena queries are slow due to many small objects. Which remediation is most effective?

  1. Add a random prefix to every object key
  2. Enable S3 Transfer Acceleration on the bucket
  3. Reduce the Athena query result location retention
  4. Compact small files into larger Parquet objects using a batch job Correct answer.

Explanation

Many small files create excessive per-object overhead for query engines, so compaction into fewer, larger columnar files improves scan efficiency and reduces cost. Transfer acceleration, key prefixing, and result retention do not address the small-file problem.

Question 16Data Engineering

Which feature of AWS Glue allows a job to process only data added since the previous run by tracking previously processed objects in S3?

  1. Trigger schedules
  2. Crawler classifiers
  3. Job bookmarks Correct answer.
  4. Development endpoints

Explanation

Glue job bookmarks persist state between runs so a job can skip objects or records already processed. Classifiers infer schemas, triggers start jobs, and development endpoints are for interactive authoring.

Question 17Data Engineering

A team needs to ingest data from hundreds of IoT devices into AWS with durable ordered delivery per device and the ability to replay records for reprocessing. Which service best meets these requirements?

  1. Amazon Kinesis Data Streams Correct answer.
  2. Amazon SNS
  3. Amazon Kinesis Data Firehose
  4. Amazon SQS standard queue

Explanation

Kinesis Data Streams retains records for a configurable period and allows consumers to replay from a chosen position, while ordering per partition key preserves device-level sequence. SQS standard queues do not guarantee ordering and delete messages after consumption, and Firehose does not support replay.

Question 18Data Engineering

A Redshift table is frequently joined on a customer_id column, and joins require substantial data movement between nodes. Which distribution style most directly colocates matching rows on the same node?

  1. DISTSTYLE AUTO with no sort key
  2. DISTSTYLE KEY on customer_id Correct answer.
  3. DISTSTYLE ALL
  4. DISTSTYLE EVEN

Explanation

Key distribution hashes the chosen column so rows with the same customer_id land on the same node, allowing collocated joins and reducing network traffic. Even distribution spreads rows round-robin, and ALL replicates the entire table to every node.

Question 19Data Engineering

Which AWS service should be used to orchestrate a multi-step ETL workflow with dependencies, retries, and scheduled runs across Glue jobs and Lambda functions?

  1. AWS Step Functions Correct answer.
  2. Amazon CloudWatch Logs
  3. Amazon Kinesis Data Firehose
  4. AWS Glue DataBrew

Explanation

Step Functions expresses workflows as state machines with dependencies, retries, and error handling, and it can invoke Glue jobs and Lambda functions. CloudWatch Logs stores logs, DataBrew prepares data, and Firehose delivers streams.

Question 20Data Engineering

A company must encrypt data at rest in Amazon S3 and wants AWS to manage the keys while still auditing key usage separately from other services. Which option best fits?

  1. SSE-KMS with a customer managed key Correct answer.
  2. SSE-S3 with Amazon S3 managed keys
  3. Client-side encryption with a key stored in application code
  4. SSE-C with keys supplied on each request

Explanation

SSE-KMS with a customer managed key gives AWS-managed encryption plus separate key policies and CloudTrail auditing of key usage. SSE-S3 uses S3-managed keys without separate key policy control, while client-side and SSE-C put key management burden on the customer.

Question 21Data Engineering

A data engineer is designing a Kinesis Data Analytics application that must handle late-arriving events and produce accurate windowed aggregates. Which approach best addresses late data?

  1. Use processing-time windows and ignore event timestamps
  2. Use event-time processing with allowed lateness and watermarks Correct answer.
  3. Disable checkpointing to avoid waiting for late records
  4. Increase the number of shards to reduce lateness

Explanation

Event-time processing with watermarks and an allowed-lateness setting lets the application wait briefly for out-of-order records while still producing timely results. Processing-time windows ignore the actual event time and misattribute late records, and shard count or checkpointing do not solve lateness.

Question 22Data Engineering

A Glue ETL job reading from S3 shows many small input partitions and long shuffle times. Which combination of changes most improves performance?

  1. Move the data to DynamoDB and run the job against that table
  2. Reduce the number of DPUs and disable compression
  3. Convert output to uncompressed CSV and add more partitions
  4. Increase the number of DPUs and coalesce output partitions before writing Correct answer.

Explanation

Adding DPUs provides more compute for shuffles, and coalescing output partitions reduces the number of small files and task overhead. Reducing compute, removing compression, or relocating to DynamoDB would worsen scan and shuffle costs.

Question 23Data Engineering

An organization needs near-real-time analytics on streaming data with SQL, joining a stream against a reference table in S3 that updates hourly. Which design is most appropriate on AWS?

  1. Kinesis Data Analytics with the reference table refreshed periodically in the application Correct answer.
  2. Athena queries issued every second against raw stream objects in S3
  3. Glue batch jobs scheduled every hour over the raw stream
  4. Redshift COPY run every minute from Firehose delivery files

Explanation

Kinesis Data Analytics (Apache Flink) supports continuous SQL over streams and can periodically reload reference data for enrichment. Athena, Redshift COPY, and hourly Glue batches operate on stored data and cannot deliver near-real-time stream processing.

Question 24Data Engineering

A Redshift workload mixes short dashboard queries with long-running transformations. The team wants to prevent the transformations from monopolizing cluster resources. Which feature should be used?

  1. A larger single-node cluster
  2. Cross-region snapshot copies
  3. Workload management queues with query monitoring rules Correct answer.
  4. Manual VACUUM during peak hours

Explanation

Redshift workload management lets you route queries into separate queues with memory and concurrency allocations, and query monitoring rules can abort or log runaway queries. A bigger node, peak-hour vacuuming, or snapshot copies do not isolate workloads.

Question 25Data Engineering

A data lake must support ACID transactions, time travel, and schema evolution for tables queried by both Athena and Spark. Which approach best satisfies these requirements?

  1. Store data as uncompressed JSON with Hive-style partitions
  2. Store all tables in a single Redshift cluster and query via Spectrum only
  3. Use S3 versioning alone to provide transactional guarantees
  4. Use an open table format such as Apache Iceberg registered in the Glue Data Catalog Correct answer.

Explanation

Open table formats like Iceberg provide ACID commits, snapshot-based time travel, and schema evolution, and they integrate with Athena and Spark through the Glue Data Catalog. Plain JSON, S3 versioning, or a single Redshift cluster do not deliver table-level ACID semantics for a shared data lake.

Question 26Data Engineering

A data engineer needs to catalog data stored in Amazon S3 so that Amazon Athena can query it. Which AWS service should be used to create and manage the table definitions?

  1. AWS Glue Data Catalog Correct answer.
  2. AWS Lake Formation
  3. Amazon Redshift Spectrum
  4. Amazon EMR

Explanation

The AWS Glue Data Catalog is a centralized metadata repository that stores table definitions, schemas, and partition information. Athena uses the Data Catalog to discover and query S3 data. Lake Formation manages permissions but relies on the Data Catalog for metadata.

Question 27Data Engineering

A company wants to move data from an on-premises PostgreSQL database to Amazon S3 every hour with minimal code. Which AWS service is designed for this task?

  1. Amazon Kinesis Data Firehose
  2. AWS DataSync Correct answer.
  3. AWS Glue
  4. AWS Database Migration Service (AWS DMS)

Explanation

AWS DataSync is an online data transfer service that automates and accelerates moving data between on-premises storage and AWS storage services such as S3. It is designed for scheduled, incremental transfers with minimal operational overhead.

Question 28Data Engineering

A data engineer is using AWS Glue to transform data. The job must process only new data since the last run. Which Glue feature should be used?

  1. Glue crawler
  2. Job bookmarks Correct answer.
  3. Glue DataBrew
  4. Glue triggers

Explanation

AWS Glue job bookmarks track the state of data processed by a job, allowing it to process only new data on subsequent runs. This avoids reprocessing old data and reduces cost.

Question 29Data Engineering

Which AWS service provides a fully managed, petabyte-scale data warehouse that uses columnar storage and massively parallel processing?

  1. Amazon Athena
  2. Amazon Redshift Correct answer.
  3. Amazon RDS
  4. Amazon DynamoDB

Explanation

Amazon Redshift is a fully managed data warehouse service that uses columnar storage and massively parallel processing to run complex analytical queries on large datasets. It is optimized for petabyte-scale workloads.

Question 30Data Engineering

A data engineer is configuring an Amazon Kinesis Data Firehose delivery stream to Amazon S3. The data must be compressed to reduce storage costs. Which compression format is supported by Firehose?

  1. BZIP2
  2. GZIP Correct answer.
  3. LZ4
  4. Snappy

Explanation

Amazon Kinesis Data Firehose supports GZIP, ZIP, and Snappy compression for data delivered to Amazon S3. GZIP is a valid and commonly used option for reducing storage size.

Question 31Data Engineering

A company stores sensitive data in Amazon S3 and must encrypt it at rest. They want AWS to manage the encryption keys and also need an audit trail of when and by whom the keys were used. Which solution meets these requirements?

  1. SSE-KMS with a customer managed key Correct answer.
  2. SSE-C with client-provided keys
  3. SSE-S3 with S3 Bucket Keys
  4. Client-side encryption with a custom key store

Explanation

SSE-KMS uses AWS KMS to manage keys and provides an audit trail through AWS CloudTrail, recording key usage. A customer managed key gives control over key policies and enables separate auditing.

Question 32Data Engineering

A data engineer is building an AWS Glue ETL job that reads from Amazon S3 and writes to Amazon Redshift. The job must handle schema changes in the source data without failing. Which Glue feature should be enabled?

  1. Crawler classifiers
  2. DynamicFrame schema evolution Correct answer.
  3. Glue DataBrew recipes
  4. Job bookmarks

Explanation

DynamicFrames in AWS Glue can handle schema evolution by reconciling differences and adding missing columns, allowing jobs to continue when source schemas change. This prevents job failures due to unexpected new fields.

Question 33Data Engineering

A data engineer needs to run a SQL query that joins a large Amazon Redshift table with a smaller dimension table. To optimize performance, which distribution style should be used for the smaller table?

  1. KEY
  2. EVEN
  3. AUTO
  4. ALL Correct answer.

Explanation

The ALL distribution style replicates the entire smaller table to every compute node, eliminating the need to redistribute data during joins. This is ideal for dimension tables that are relatively small and frequently joined.

Question 34Data Engineering

A company uses AWS Step Functions to orchestrate an ETL workflow. The workflow must retry a failed AWS Glue job up to three times with exponential backoff. Which Step Functions feature should be used?

  1. Wait state
  2. Map state
  3. Retry and Catch fields Correct answer.
  4. Parallel state

Explanation

Step Functions supports Retry and Catch fields on states to handle errors and implement retry logic with configurable intervals and backoff. This allows automatic retries of failed tasks such as Glue jobs.

Question 35Data Engineering

A data engineer is designing a solution to ingest streaming data from thousands of sources into Amazon S3 with automatic scaling and no server management. Which service should be used?

  1. Amazon MSK
  2. AWS Glue Streaming ETL
  3. Amazon Kinesis Data Streams
  4. Amazon Kinesis Data Firehose Correct answer.

Explanation

Kinesis Data Firehose is a fully managed service that automatically scales to handle streaming data and delivers it to destinations like S3 without server management. It requires no capacity provisioning.

Question 36Data Engineering

A data engineer must ensure that data loaded into Amazon Redshift from Amazon S3 is compressed and encrypted. Which Redshift feature should be used for efficient parallel loading?

  1. Redshift Spectrum
  2. AWS Glue DataBrew
  3. COPY command with GZIP and MANIFEST Correct answer.
  4. COPY command with COMPUPDATE

Explanation

The Redshift COPY command can load compressed (e.g., GZIP) and encrypted data from S3 in parallel. Using a manifest ensures that all files are loaded correctly and can improve performance.

Question 37Data Engineering

A data engineer is using AWS Glue to crawl an Amazon S3 bucket. The data is stored in nested JSON format. Which Glue crawler classifier should be used to infer the schema?

  1. JSON classifier Correct answer.
  2. XML classifier
  3. Grok classifier
  4. CSV classifier

Explanation

A JSON classifier is designed to parse JSON data and infer its schema, including nested structures. It is the appropriate choice for nested JSON files.

Question 38Data Engineering

A company needs to analyze clickstream data in real time using SQL. The data is ingested into Amazon Kinesis Data Streams. Which service should be used to run continuous SQL queries on the stream?

  1. Amazon Athena
  2. Amazon EMR
  3. Amazon Kinesis Data Analytics Correct answer.
  4. AWS Glue

Explanation

Amazon Kinesis Data Analytics for SQL applications allows you to run continuous SQL queries on streaming data from Kinesis Data Streams. It processes and analyzes data in real time.

Question 39Data Engineering

A data engineer is designing a data lake on Amazon S3. The data must be encrypted at rest, and access must be audited. The engineer wants to use AWS managed keys but also needs to track key usage separately. Which approach meets these requirements?

  1. SSE-KMS with a customer managed key Correct answer.
  2. SSE-KMS with an AWS managed key
  3. Client-side encryption with a custom key
  4. SSE-S3 with default encryption

Explanation

SSE-KMS with a customer managed key provides separate auditing of key usage through CloudTrail and allows control over key policies. AWS managed keys can be used but do not provide the same level of audit separation.

Question 40Data Engineering

A data engineer is optimizing an AWS Glue job that reads many small files from Amazon S3. The job is slow due to excessive task overhead. Which technique should be used to improve performance?

  1. Use groupFiles to combine small files Correct answer.
  2. Increase the number of DPUs
  3. Use a larger instance type
  4. Enable job bookmarks

Explanation

Grouping small files into larger ones reduces task overhead and improves read performance in Glue. The groupFiles option in the Glue DynamicFrame reader can coalesce small files.

Question 41Exploratory Data Analysis

A data scientist wants a quick numeric summary of every feature in a pandas DataFrame, including count, mean, standard deviation, and quartiles, before building any model. Which single call produces this summary for the numeric columns?

  1. df.info()
  2. df.describe() Correct answer.
  3. df.value_counts()
  4. df.corr()

Explanation

The describe method computes count, mean, standard deviation, minimum, quartiles, and maximum for numeric columns by default, giving a compact distributional overview. It is the standard first-pass summary for spotting scale differences and outliers.

Question 42Exploratory Data Analysis

An analyst loads a CSV and wants to know how many missing values exist in each column. Which expression returns a per-column count of nulls?

  1. df.isnull().sum() Correct answer.
  2. df.fillna(0).count()
  3. df.duplicated().sum()
  4. df.dropna().shape

Explanation

isnull produces a boolean frame marking missing cells, and summing down each column counts the True entries, yielding missing counts per column. This is the standard profiling step before deciding on imputation strategy.

Question 43Exploratory Data Analysis

A histogram of a feature shows a long right tail with most observations clustered near zero. Which statement best characterizes this distribution?

  1. It is bimodal, so the mean is undefined.
  2. It is left-skewed, so the mean is below the median.
  3. It is right-skewed, so the mean exceeds the median. Correct answer.
  4. It is symmetric, so mean and median are approximately equal.

Explanation

A long tail extending toward high values pulls the mean upward relative to the median, which is the defining pattern of right skew. Recognizing this guides transforms such as a log or Box-Cox before modeling.

Question 44Exploratory Data Analysis

A modeler computes a Pearson correlation matrix and sees a coefficient of 0.02 between two numeric features. What is the most defensible interpretation?

  1. One feature causes the other to be constant.
  2. The features are strongly related nonlinearly.
  3. There is little linear association between the two features. Correct answer.
  4. The two features are identical in distribution.

Explanation

Pearson correlation measures only the strength of linear association, so a value near zero indicates essentially no linear relationship. A nonlinear dependency could still exist, but the coefficient alone does not establish it.

Question 45Exploratory Data Analysis

Which visualization is most appropriate for comparing the distribution of a single numeric feature across several categorical groups?

  1. A scatter plot of two numeric features
  2. A heat map of pairwise correlations
  3. A box plot grouped by category Correct answer.
  4. A pie chart of category totals

Explanation

Grouped box plots display median, quartiles, and outliers for each category side by side, making distributional comparisons direct. This supports decisions about per-group scaling or stratified splits.

Question 46Exploratory Data Analysis

A dataset has a feature whose values range from 1 to 1,000,000 while all other features range between 0 and 1. Which exploratory finding should be flagged before training a distance-based model?

  1. Perfect multicollinearity
  2. Duplicate rows in the dataset
  3. Severe scale disparity across features Correct answer.
  4. Severe class imbalance in the target

Explanation

Features on wildly different scales dominate distance and gradient computations, so standardization or normalization is needed. Exploratory range checks expose this before it silently degrades model behavior.

Question 47Exploratory Data Analysis

What does a quartile-quartile plot of a feature against a normal distribution primarily help an analyst assess?

  1. How many rows contain missing values
  2. How closely the feature follows a normal distribution Correct answer.
  3. Whether the feature has duplicate values
  4. Whether the target is balanced across classes

Explanation

A Q-Q plot compares sample quantiles with theoretical normal quantiles, and systematic departures from the diagonal indicate skew or heavy tails. This informs whether transformations or robust methods are warranted.

Question 48Exploratory Data Analysis

During profiling, an analyst finds that a numeric column stores values like 101, 102, 103 that actually encode three customer segments. What data-quality issue does this represent?

  1. A time-series seasonality pattern
  2. A categorical variable encoded as numbers Correct answer.
  3. A continuous variable with high kurtosis
  4. A target leakage problem

Explanation

Numeric codes standing for categories can mislead algorithms into treating them as ordered magnitudes. Exploratory checks on the number of distinct values reveal such disguised categoricals so they can be one-hot encoded or otherwise handled.

Question 49Exploratory Data Analysis

A feature has 40 percent missing values, and inspection shows the missingness occurs far more often for one target class than the other. Which conclusion is most appropriate?

  1. The missingness is informative and may itself be a useful signal. Correct answer.
  2. The feature should be dropped without further review.
  3. The values are missing completely at random, so deletion is safe.
  4. The missing values can be ignored because 40 percent is below half.

Explanation

When missingness correlates with the target, it is not missing completely at random and can carry predictive information. Analysts should consider a missing-indicator feature rather than silently deleting or ignoring the gaps.

Question 50Exploratory Data Analysis

An analyst wants to detect outliers in a roughly symmetric numeric feature using the interquartile range rule. Which threshold correctly defines the fences?

  1. The mean plus or minus one standard deviation
  2. The 1st and 99th percentiles
  3. Q1 minus 1.5 times IQR and Q3 plus 1.5 times IQR Correct answer.
  4. The minimum and maximum observed values

Explanation

The standard Tukey rule places fences at 1.5 interquartile ranges below the first quartile and above the third quartile. Points beyond the fences are flagged as potential outliers for review.