CompTIA Data+ Practice Test (DA0-002)

The correct answer is File storage. This scenario describes a classic file storage system, often implemented as Network Attached Storage (NAS). Key characteristics include its hierarchical organization (data stored in files within a tree of folders/directories) and its ability to provide shared access to multiple users over a network. The description of a 'mapped network drive' and a 'folder-and-file structure' directly aligns with the definition of file storage.

• Object storage is incorrect because it stores data in a flat structure of 'buckets' or 'containers', not a hierarchical file system. Access is typically handled via APIs rather than a browsable file system.
• Block storage is incorrect because it operates at a lower level, breaking data into fixed-size blocks. A file system is built on top of block storage, but block storage itself does not inherently present a folder/file structure to the end-user.
• Local storage is incorrect because it refers to storage directly attached to or inside a single user's computer (e.g., a local C: drive). The scenario specifies a 'central server' and shared access for the entire team, which points to a networked, not local, solution.

The correct answer is Exploding. The 'exploding' transformation is used to convert a single row that contains a list-like structure (such as a comma-separated string or an array) into multiple rows, one for each element in the list. In this scenario, it would create a new row for each genre associated with a movie, allowing for accurate aggregation by genre.

• Parsing involves analyzing a string of text to break it into smaller, meaningful components. While parsing the 'genres' string is a necessary step to separate the genres, the overall technique that creates new rows from these components is known as exploding.
• Binning is a technique used to group a range of continuous numerical values into discrete intervals or 'bins'. It is not suitable for handling list-like categorical data.
• Imputation is the process of replacing missing values in a dataset with substituted values. This is not relevant to the task, as the problem is about restructuring existing data, not filling in missing data.

Because the request is time-sensitive, one-off, and outside the scope of the recurring overnight dashboard, the analyst should build and share an ad hoc dashboard. Ad hoc dashboards (or reports) are created on demand to answer a specific question and are not tied to a fixed refresh schedule. Waiting for the nightly recurring dashboard, sending a weekly static report, or engineering a continuous real-time stream would all miss the immediate noon deadline or require unnecessary effort for a single meeting.

The correct answer is Incompleteness. Data completeness is a data quality dimension that measures whether all necessary data is present in a dataset. In this scenario, the postal_code is a required field for the direct mail campaign, and its absence in 30% of records makes the dataset incomplete for the intended business purpose.

• Duplication is incorrect because it refers to the same record appearing multiple times in a dataset. The problem described is missing data within records, not repeated records.
• Redundancy is incorrect as it refers to the same data being stored in multiple places, which can lead to inconsistencies but is not the issue here.
• Outliers are incorrect because they are data points that are present but deviate significantly from the other observations. The issue identified is the absence of data (null values), not the presence of abnormal data.

The correct answer is Anonymization. Anonymization is the process of removing or modifying personally identifiable information (PII) and personal health information (PHI) to ensure that the individuals who are the subjects of the data cannot be identified. This technique is essential when sharing sensitive data for external research or statistical analysis, as it protects privacy while maintaining the analytical value of the data.

• Data masking is a plausible but less appropriate choice. Masking typically replaces sensitive data with realistic but fabricated data and is most often used for internal purposes like software testing or training where the data format must be preserved. While it is a form of data obfuscation, anonymization is the more precise term for the goal of making re-identification impossible for research purposes.
• Encryption at rest protects data that is stored on a disk or in a database by making it unreadable without a decryption key. It does not protect the data once it has been decrypted and shared with the third-party researchers.
• Role-based access control (RBAC) is a method for managing who can access data within a system based on their job function. It controls access permissions but does not alter the data itself to make it safe for sharing with an external entity.

Robust scaling uses the median as its center and the interquartile range (IQR) as its scale factor. Because both the median and IQR are resistant to extreme values, the transformation keeps most observations on a comparable scale while preventing a handful of very large lifetime_value records from stretching the range of the whole variable.

Min-max scaling rescales the data between fixed bounds such as 0 and 1, but the presence of even a single extreme value pushes all other observations into a narrow interval, so outliers still dominate. Standardization (Z-score) centers data on the mean and scales by the standard deviation; since both statistics are sensitive to outliers, the resulting values can still be overstretched by extreme cases. A log transform followed by min-max scaling can reduce skew but will still tie the upper bound of the scale to the largest remaining value, offering less protection than an IQR-based approach. Therefore, robust scaling is the most appropriate choice for this scenario.

Jurisdictional requirements focus on laws that apply because of where data is collected, stored, or accessed. Under the GDPR, Singapore is a "third country," so exporting EU personal data there is lawful only if an approved cross-border transfer mechanism-such as an adequacy decision, Standard Contractual Clauses, or Binding Corporate Rules-has been validated. Verifying and implementing that legal basis directly addresses the geographic (jurisdictional) aspect of compliance. The other options deal with technical safeguards (encryption), disaster-recovery timing, or metadata maintenance; none of those, by themselves, satisfy location-based legal obligations.

The correct answer is a large language model (LLM). LLMs are a type of artificial intelligence specifically designed to understand, process, and generate human-like text based on the vast amounts of data they are trained on. This makes them ideal for tasks like summarizing unstructured text, analyzing sentiment, and identifying themes from customer reviews.

• Robotic process automation (RPA) is incorrect because it is designed to automate repetitive, rule-based tasks by mimicking human interactions with digital systems, such as data entry or file transfers. It does not have the capability to interpret unstructured text and generate novel summaries.
• A dimensional table is a data warehouse structure that stores descriptive attributes to provide context to facts in a fact table. It is a way to organize structured data for analysis, not a tool for processing and summarizing unstructured text.
• While an LLM is a type of foundational model, 'foundational model' is a broader category that can include models for images, audio, and other data types, not just text. 'Large language model' is the more specific and therefore most appropriate choice for a task that exclusively involves processing and generating text.

Because the missing values occur in the target variable-not in the predictor features-the rows cannot contribute to supervised learning. Imputing or transforming the missing target would inject fabricated labels and risk corrupting the model. Binning or scaling features does nothing to resolve the missing label, and the library will still fail. Given that the affected subset represents only 0.52 % of the data and its removal does not bias the class distribution, listwise deletion (dropping those rows) is the proper cleansing step. Imputing the mode would create false churn outcomes; scaling features leaves the NULLs untouched; and binning the target is impossible without a value to bin, so those choices are incorrect.

Unattended bots are designed to run end-to-end automations without any human trigger. They can be scheduled in an orchestrator or control room to start at a specific time, log in to systems, execute every step of the workflow, and distribute the finished report. Attended bots require a user to start them and therefore cannot guarantee execution at 03:00 when staff are offline. Citizen-developer desktop recordings are intended for design or personal productivity, not for lights-out production runs. Test or QA bots are used for validation in non-production environments and are not scheduled in production for routine reports. Because the scenario calls for lights-out, time-based execution, the unattended, scheduler-driven bot is the only suitable choice.

Under GDPR, personal data may only be transferred outside the European Economic Area if the controller applies an approved safeguard such as Standard Contractual Clauses (SCCs) or Binding Corporate Rules. This directly addresses the cross-border replication that the auditor flagged. GDPR's storage-limitation principle also requires organisations to delete or anonymise personal data when it is no longer needed, so a documented retention schedule for snapshots is necessary. Restricting or geo-fencing replication to approved regions ensures ongoing jurisdictional compliance while still permitting a disaster-recovery copy. The distractors focus on security controls (encryption, masking), PCI-specific requirements, or testing activities; none of those controls by themselves fix both the unlawful transfer and the indefinite retention issues that triggered the compliance finding.

JSON is a text-based format that can nest objects and arrays inside other objects. When a JSON document includes an array-such as line_items-the structure becomes hierarchical. Relational tables, however, are two-dimensional and expect one scalar value per column, so the nested content must be "flattened" (for example, exploded into separate rows or moved to a child table) before it can be inserted.

The other options describe problems that do not inherently apply to .json files:

• JSON files are plain text; GZIP or any other compression is optional, not mandatory.
• Numbers in JSON are encoded as character strings, not as binary integers, so byte-order (endianness) translation is unnecessary.
• The JSON standard requires UTF-8 encoding for open data exchange, so full Unicode character sets are already supported.

Therefore, the most likely obstacle is flattening the nested arrays and objects into a flat relational structure.

Block storage divides data into fixed-size chunks (blocks) that are addressed directly by the operating system using low-level protocols such as iSCSI or NVMe/TCP. Because the host treats the volume like a locally attached disk, the database can format the device with its own file system and issue random reads and writes with sub-millisecond latency-ideal for OLTP workloads.

The other statements describe capabilities of other storage types:

• Object storage keeps data as immutable objects with rich metadata and often uses replication or erasure coding across regions to maximize durability; this design trades latency for scalability and resilience.
• File storage exposes a hierarchical directory structure over protocols like NFS or SMB for shared user access, which adds additional overhead compared with raw block devices.
• Erasure-coded, capacity-optimized architectures are tuned for archival or infrequently accessed data, not latency-sensitive transaction processing.

The correct answer is a fact table. In a dimensional model like a star schema, the fact table is the central table that stores quantitative measurements or metrics about a business process. In this scenario, the quantity_sold and sale_amount are the facts. The fact table also contains foreign keys that connect to the primary keys of the surrounding dimensional tables (Date, Product, and Store), which provide context to the facts.

• A dimensional table stores the descriptive attributes related to the facts, such as product names, store addresses, or customer details, not the quantitative transactional data itself.
• A bridge table is a specific type of table used to resolve many-to-many relationships between a fact and a dimension or between two dimensions, which is not the primary purpose described in the scenario.
• A staging table is a temporary storage area used during the ETL (Extract, Transform, Load) process to clean and prepare data before it is loaded into the final destination, like a data warehouse. It is not a permanent part of the analytical schema.

The correct answer is deep learning. Deep learning is a subset of machine learning that uses multi-layered neural networks to automatically learn and identify complex patterns from large amounts of raw, unstructured data, such as images. It is the primary technology behind modern image recognition and computer vision systems.

• Robotic Process Automation (RPA) is incorrect because it is designed to automate structured, rule-based tasks by mimicking human interactions with software interfaces, not to analyze or interpret unstructured data like images.
• Natural Language Processing (NLP) is incorrect as it focuses on the interaction between computers and human language (text and speech), not visual data.
• Automated reporting is incorrect because it is the process of generating reports, usually from structured data sources, and does not involve the complex analysis of image content.

Because the managers need the same metrics on a predictable cadence (every Monday) and the source data is already finalized overnight, the most efficient solution is to set up an automated, recurring dashboard refresh that runs early Monday morning. A scheduled refresh guarantees the dashboard opens with fresh data and eliminates the need for users to trigger an update. Real-time DirectQuery is unnecessary for once-per-week consumption and can add gateway overhead; an ad-hoc template still requires each manager to generate the report manually; exporting a static PDF on Friday would leave three days of weekend sales missing by Monday.

Persisting the subquery result in a session-scoped temporary table means the expensive aggregation runs only once. The main query can then join to the smaller, indexed #recent_orders table, eliminating two extra scans of the 50-million-row Orders table and sharply reducing logical reads. A CTE is usually inlined, so the optimizer may still re-evaluate it; index or lock hints leave the multiple scans in place and merely influence access or concurrency methods, so they do not address the root cause.

Dimensional modeling guidelines recommend assigning each dimension row a meaningless, sequential surrogate key-typically a 4-byte integer-rather than relying on natural business keys or wide composite keys. A compact surrogate key keeps the fact table's foreign-key columns small (saving space and index overhead), speeds join performance, shields the warehouse from changes to source-system codes, and is mandatory when tracking slowly changing dimension history. Natural keys, GUIDs and composite keys all consume more space, may change unexpectedly, and complicate surrogate-key lookups for historical versioning, so they are not preferred choices for a dimension table in a star schema.

The correct answer is the 5-point Likert scale question. This type of question asks respondents to rate their level of agreement or satisfaction on a scale, which generates ordinal data. This data is commonly treated as interval data in business analytics, allowing for the calculation of an average (mean) score. This directly meets the project's goal of calculating an average satisfaction score to track over time.

The multiple-choice option provides nominal (categorical) data. It is not possible to calculate a meaningful average from non-ordered categories like 'Easy to use' or 'Buggy'. An open-ended question generates qualitative (text) data, which is valuable for deep insights but cannot be used to calculate a numerical average without significant extra processing like sentiment analysis or manual coding. A dichotomous question provides binary ('Yes'/'No') data. While you can calculate the percentage of 'Yes' responses, it does not provide a granular score that can be averaged to show varying levels of satisfaction.

The correct answer is Predictive analysis. The scenario requires forecasting a future event, specifically, which customers are likely to cancel their subscriptions. Predictive analytics uses historical data to find patterns and make predictions about future outcomes.

• Descriptive analytics is incorrect because it focuses on summarizing past data to understand what has already happened (e.g., calculating the average subscription length of customers who have already churned). It does not forecast future events.
• Prescriptive analytics is incorrect because it recommends actions to take. While the ultimate goal is to act (the retention offer), the analyst's primary and immediate task is to identify the at-risk customers (prediction), which is a necessary step before a specific action can be prescribed.
• Inferential analytics is incorrect because it involves using a data sample to make generalizations or test hypotheses about a larger population (e.g., testing if there is a statistically significant difference in churn rates between two customer segments). The request is to create a model to score individual customers, not to test a general population hypothesis.