IsacaData Science Fundamentals Certificate Questions and Answers
A requirement states that a dashboard must load "quickly." Which requirement quality characteristic does this statement MOST clearly fail to meet?
Options:
Unambiguous
Traceable
Modifiable
Answer:
AExplanation:
Well-formed requirements share a set of quality characteristics, including being correct, complete, consistent, unambiguous, testable, traceable, and modifiable. An unambiguous requirement has only one possible interpretation, so every stakeholder, developer, and tester understands it identically. The word "quickly" is subjective: a business sponsor might expect a two-second response, an analyst might accept ten seconds, and an engineer might consider a minute acceptable for a large query. Because the statement can be read in several ways, it fails the unambiguous standard most directly. The usual remedy is to replace the vague term with a measurable criterion, for example that the dashboard must display within three seconds for ninety-five percent of requests under normal load. Doing so also makes the requirement testable, a related characteristic that the vague wording violates as a secondary consequence. Traceability, option B, concerns whether a requirement can be linked back to its business need and forward to design, development, and test artifacts; the wording of the statement does not affect it. Modifiability, option C, concerns whether the requirement can be changed without disrupting others and likewise is unrelated to vagueness. Precise requirements reduce rework and disputes during delivery of data science projects. Reference: Data Science Fundamentals Study Guide – Data Science Process: Defining Data Requirements and Quality Characteristics.
===============
Which of the following data collection methods is grounded in facts and numbers, and removes personal bias from the process?
Options:
Qualitative
Exploratory
Quantitative
Answer:
CExplanation:
Quantitative data collection gathers numerical information that can be counted or measured, through instruments such as structured surveys with closed-ended questions, sensors, transaction logs, and controlled experiments. Because the results are expressed as numbers, they can be summarized, compared, and tested with statistical procedures, and the process is standardized so that other researchers can repeat it and obtain consistent results. Standardization reduces the influence of the collector's personal interpretation, which is why the method is regarded as objective and grounded in facts and figures. Option C is correct. Qualitative methods, option A, such as interviews, focus groups, and open-ended observation, gather descriptive, non-numerical information about opinions, motivations, and experiences. They deliver depth and context but depend more on the judgment of the researcher in recording and interpreting responses, so subjectivity is harder to eliminate. Exploratory, option B, describes the purpose of a study aimed at discovering patterns or generating hypotheses and may use either type of data; it is not a data collection method defined by numerical measurement. No method is entirely free of bias, since sampling and instrument design can introduce it, but quantitative approaches minimize interpretive bias. Many projects combine both approaches to pair measured results with explanatory insight. Reference: Data Science Fundamentals Study Guide – Data Science Concepts: Data Collection Methods (Quantitative and Qualitative).
===============
What is the MAIN distinction between clustering and classification?
Options:
Clustering is unsupervised, while classification uses predetermined groupings.
Classification is performed as an automated real-time process, while clustering is performed manually.
Clustering is defined by analysts, while classification uses an algorithm.
Answer:
AExplanation:
The defining difference lies in whether the categories are known before the algorithm runs. Classification is a supervised technique: a model is trained on records that already carry labels, learns the features associated with each class, and then assigns new observations to one of those predetermined groups. Common algorithms include decision trees, logistic regression, and naive Bayes. Clustering is unsupervised. It receives unlabeled data and discovers natural groupings on its own by measuring similarity or distance between observations, as in k-means or hierarchical clustering; the resulting groups are interpreted afterward rather than specified in advance. Option A states this contrast accurately. Option B introduces a false dichotomy between real-time automation and manual processing; both methods are algorithmic, and either can operate in batch or streaming contexts. Option C reverses the roles: it is classification in which analysts or domain experts define the categories through labeling, whereas clustering relies on the algorithm to propose the structure, even though an analyst may choose parameters such as the number of clusters. Recognizing the supervised versus unsupervised distinction is essential for selecting an appropriate technique when labeled outcomes either exist or do not. Reference: Data Science Fundamentals Study Guide – Data Science Concepts: Machine Learning Approaches (Supervised and Unsupervised Learning).
===============
A model achieves near-perfect results on the training data but performs poorly when applied to new, unseen data. This behavior is MOST likely an example of:
Options:
Underfitting
Overfitting
Normalization
Answer:
BExplanation:
Overfitting arises when a model learns the training data too closely, absorbing random noise and idiosyncratic details in addition to the genuine underlying pattern. The model therefore achieves very low error on the data it was trained on but generalizes badly, because the quirks it memorized do not recur in new observations. The telltale sign is a wide gap between training performance and validation or test performance. Overfitting is encouraged by excessively complex models, too many features relative to the number of observations, and insufficient data. Remedies include collecting more data, simplifying the model, selecting fewer features, applying regularization, pruning decision trees, stopping training early, and evaluating with cross-validation. Underfitting, option A, is the opposite problem: a model too simple to capture the structure of the data performs poorly on both training and new data. Normalization, option C, is a preprocessing step that rescales feature values to a common range; it is not a model behavior and does not explain the pattern described. Understanding the bias and variance trade-off is central here, since overfit models exhibit high variance while underfit models exhibit high bias, and practitioners seek the balance that minimizes error on unseen data. Reference: Data Science Fundamentals Study Guide – Data Science Concepts: Model Evaluation, Overfitting and Underfitting.
===============
Moore's law states that:
Options:
capacity to process big data decreases by a factor of two every two years.
computer memory increases by a factor of two, but processing is halved.
computing power increases by a factor of two every two years.
Answer:
CExplanation:
Moore's law originates from an observation by Gordon Moore, later a co-founder of Intel, that the number of transistors that can be placed on an integrated circuit doubles at a regular interval, commonly stated as every two years. Because transistor density drives processing capability, the law is generally summarized as computing power increasing by a factor of two roughly every two years, with cost per unit of computation falling in parallel. Its relevance to data science is direct: sustained growth in processing speed and storage economics made it feasible to collect, store, and analyze data at the scale associated with big data, and it made computationally demanding techniques such as deep neural networks practical. Options A and B each invert the principle. Option A claims that processing capacity declines, which contradicts the entire trend, and option B pairs growing memory with halving processing, a relationship Moore never described. Candidates should also keep in mind that the law is an empirical observation rather than a physical law, and that its pace has slowed as fabrication approaches atomic limits, which is one reason distributed and parallel computing have become central to modern analytics. Reference: Data Science Fundamentals Study Guide – Data Science Concepts: Drivers of Data Science, Big Data and Computing Power.
===============
Which of the following should be understood as compulsory restrictions that must be followed?
Options:
Codes of conduct
Ethics
Policies
Answer:
CExplanation:
Governance frameworks distinguish between instruments that compel behavior and those that guide it. A policy is a formal statement approved by management that establishes what personnel and systems must or must not do. Policies carry the authority of the organization and are accompanied by consequences, including disciplinary action, for noncompliance, and they are supported by standards, procedures, and controls that put them into practice. Data handling policies, for example, specify classification, retention, access, and sharing rules that employees are obligated to follow. For this reason policies are the compulsory restrictions in the question, and option C is correct. Ethics, option B, are principles of right and wrong that shape judgment and are not self-enforcing; they inform decisions in situations that rules do not anticipate and often extend beyond what the law or policy demands. A code of conduct, option A, articulates an organization's values and the standards of behavior it expects, and while it may be tied to employment, it generally communicates expectations and aspirations in broad terms rather than detailed mandatory requirements. Understanding where each instrument sits in the hierarchy helps data professionals determine which obligations are binding and which are advisory when handling sensitive information. Reference: Data Science Fundamentals Study Guide – Data Management: Data Governance, Policies and Ethical Guidelines.
===============
What does Bayes' Theorem state?
Options:
There are four classes of scales that are unchanging.
Processing power doubles in size every two years.
If something is true, something related to it is potentially true.
Answer:
CExplanation:
Bayes' Theorem is a result in probability theory describing how to revise the probability of a hypothesis once new evidence is observed. Expressed mathematically, the probability of A given B equals the probability of B given A multiplied by the prior probability of A, divided by the probability of B. The theorem formalizes conditional reasoning: knowledge that one event has occurred or is true changes the likelihood of a related event. This is why it appears in spam filtering, medical diagnostic testing, and the naive Bayes classifier, where prior beliefs are combined with observed features to yield a posterior probability. Option C expresses that relationship in general terms, namely that if something is true, a related event becomes potentially true. Option A confuses the theorem with the four scales of measurement (nominal, ordinal, interval, and ratio) associated with S. S. Stevens, which concern how variables are classified rather than how probabilities are updated. Option B restates Moore's law, an observation about transistor density and computing power with no connection to probability. Candidates should be able to distinguish these three named principles, since each belongs to a different strand of the data science foundation. Reference: Data Science Fundamentals Study Guide – Data Science Concepts: Probability and Statistics Fundamentals (Bayes' Theorem).
===============
Which of these roles is typically held by a director, manager, or C-level executive who bears accountability for an organization's data assets?
Options:
Data owner
Data steward
Data custodian
Answer:
AExplanation:
Data governance assigns distinct roles so that accountability for data is explicit. The data owner is a senior business leader, such as a director, manager, or executive, who is ultimately accountable for a defined set of data assets. The owner decides how the data are classified, who may access them, how long they are retained, and what uses are acceptable, and answers to the organization and to regulators for those decisions. Option A matches that description. The data steward, option B, acts on behalf of the owner at an operational level. Stewards define and monitor data quality standards, maintain business definitions and metadata, resolve data issues, and ensure that policies are followed in daily practice, and they are usually subject-matter experts rather than executives. The data custodian, option C, is a technical role, commonly within IT, responsible for implementing the controls the owner specifies: storing the data securely, managing backups, applying access permissions, and operating the underlying systems. A helpful summary is that the owner decides, the steward governs the content, and the custodian protects and maintains the infrastructure. Clear role assignment prevents gaps in which everyone assumes that someone else is responsible for the quality and security of an important data set. Reference: Data Science Fundamentals Study Guide – Data Management: Data Governance Roles and Responsibilities.
===============
Which choice BEST captures the distinction between operational and non-operational databases?
Options:
Operational databases update data in real time.
Non-operational databases process in real time.
Non-operational databases use online transaction processing (OLTP).
Answer:
AExplanation:
Operational databases support the daily running of an organization. They record transactions as they occur, such as orders, payments, reservations, and inventory movements, and they are designed for many small, fast reads and writes, with strong consistency guarantees. This workload is called online transaction processing, or OLTP, and it requires that data be updated in real time so that the current state of the business is always reflected. Non-operational databases, which include data warehouses, data marts, and archives, serve analytical and historical purposes. They are populated periodically from operational sources, hold large volumes of integrated historical data, and are tuned for complex queries and aggregation, a pattern known as online analytical processing, or OLAP. They are not updated transaction by transaction. Option A therefore expresses the distinction accurately. Option B reverses the relationship by assigning real-time processing to the non-operational type. Option C is likewise incorrect because OLTP is the defining workload of operational systems, whereas non-operational systems use OLAP. The separation protects performance, since heavy analytical queries run against a warehouse instead of competing with customer-facing transactions on the operational system. Recognizing which type of database fits a given requirement is fundamental to designing a sound data architecture. Reference: Data Science Fundamentals Study Guide – Data Management: Database Types (Operational and Non-Operational Databases, OLTP and OLAP).
===============
Which of the following is an example of a violation of the intended purpose for data collection?
Options:
A grocery delivery service asking for a customer's home address
An organization asking for the billing address associated with a credit card
A cellular provider asking for the spending habits of household occupants
Answer:
CExplanation:
Responsible data collection rests on the principle of purpose limitation: personal data should be gathered for specified, legitimate purposes and should be limited to what is necessary to fulfill them. This is closely tied to data minimization, and both are embedded in privacy regimes such as the GDPR and in organizational ethics policies. The test is whether each data element has a clear, direct connection to the service being provided. A grocery delivery company cannot deliver without a home address, so option A is justified. Billing addresses are routinely checked during card payment authorization to help detect fraud, which supports option B. A cellular provider, however, supplies voice and data connectivity, and information about the spending habits of household occupants has no bearing on that function. Collecting it signals an intention to profile or monetize customers beyond the stated reason, and it therefore represents the violation described in option C. Where such data are wanted for a secondary use, the organization would need to disclose the purpose and obtain informed consent. Evaluating the necessity of each field before it is requested reduces regulatory exposure, limits the damage of a breach, and protects trust between the organization and its customers. Reference: Data Science Fundamentals Study Guide – Data Management: Data Ethics, Privacy Principles and Purpose Limitation.
===============
Which of the following is an example of qualitative data?
Options:
Days of the week
Number of children
Gender
Answer:
CExplanation:
Qualitative, or categorical, data records a quality or attribute of an observation rather than a measured amount. Its values are labels that place each record into a group, and arithmetic on those labels carries no meaning. Gender is a clear instance: the values identify membership in a category, there is no natural numeric magnitude, and operations such as summing or averaging are meaningless. It is analyzed through frequencies, proportions, and cross-tabulations rather than means and standard deviations. Number of children, option B, is quantitative because it is a count; it is discrete, supports arithmetic, and permits a meaningful average or total. Days of the week, option A, may look categorical, but it is less definitive as an example. The labels follow a fixed order and a repeating cycle, and in analytical work they are frequently encoded numerically or treated as an ordered feature, which blurs the line that gender draws so cleanly. The question therefore asks for the unambiguous nominal attribute. Distinguishing qualitative from quantitative data, and nominal from ordinal, interval, and ratio scales, determines which summary statistics, visualizations, and models are valid for a given variable. Reference: Data Science Fundamentals Study Guide – Data Science Concepts: Data Types and Levels of Measurement.
===============
Which of these communication methods used in stakeholder presentations has the potential to be misunderstood if not applied correctly?
Options:
Kinesthetic
Audible
Visual
Answer:
CExplanation:
Presentations of analytical results typically draw on audible, visual, and kinesthetic communication. Visual methods, which include charts, graphs, tables, maps, and diagrams, are the most common and the most powerful, because the eye can absorb patterns quickly. That strength is also the source of risk. A visual encodes data through choices about chart type, axis scale, color, labeling, and scope, and each choice can alter the impression the audience takes away. Truncated axes can exaggerate differences, three-dimensional effects can distort proportions, cluttered or unlabeled elements can obscure the message, and a chart shown without context can invite conclusions the data do not support. Viewers also interpret a static image independently, with no opportunity to ask what it means, which allows several readings to coexist. Option C is therefore correct. Audible communication, option B, takes place in conversation, where the presenter can adjust wording, ask for confirmation, and answer questions immediately. Kinesthetic communication, option A, involves hands-on demonstration or interaction, so participants test their understanding directly. Effective visual reporting follows established principles: select the form that fits the data and the question, label clearly, maintain honest scales, emphasize one message, and supply a brief narrative. Reference: Data Science Fundamentals Study Guide – Data Science Process: Communicating Results to Stakeholders and Data Visualization.
===============
What is the PRIMARY distinction between a data lake and a data warehouse?
Options:
Data lakes store internal data, while data warehouses store external data.
Data lakes are structured, while data warehouses are unstructured.
Data lakes store raw data, while data warehouses store scrubbed data.
Answer:
CExplanation:
The two repositories serve different analytical needs and differ chiefly in the state of the data they hold. A data lake ingests data as it arrives from source systems, without cleansing or remodeling, and retains it in raw form. Because the structure is imposed only when the data are queried, a lake is highly flexible and supports exploratory work and machine learning on diverse formats. A data warehouse holds data that have already been extracted, cleansed, conformed, and organized into a defined schema, commonly through an ETL process, so that business users can run consistent reporting and business intelligence queries with confidence. Option C expresses this contrast: raw versus scrubbed. Option A is wrong because both repositories can contain internal and external data; origin does not separate them. Option B inverts the actual relationship. Warehouses are structured and schema-on-write, while lakes accept structured, semi-structured, and unstructured content. The distinction carries consequences for users and cost. Lakes favor data scientists and engineers who need access to original detail, whereas warehouses favor analysts who need reliable curated figures. Many organizations now combine the two in a lakehouse design that layers warehouse-style management over lake storage. Reference: Data Science Fundamentals Study Guide – Data Management: Data Storage Architectures (Data Lakes and Data Warehouses).
===============
Reporting of data results should be tailored to which of the following?
Options:
the audience's lexicon.
kinesthetic learning style.
standardized graphic templates.
Answer:
AExplanation:
The purpose of reporting is to enable stakeholders to understand findings and act on them, and that depends on how well the message fits the people receiving it. Executives, technical peers, regulators, and operational staff differ in background knowledge, vocabulary, and the decisions they face. Reports should therefore be tailored to the audience's lexicon: the terminology, level of technical detail, and framing that the audience uses and understands. A finance leader may expect results in terms of margin and return, an engineer in terms of error rates and confidence, and a general audience in plain language with jargon explained. Matching vocabulary prevents misinterpretation, builds credibility, and focuses attention on implications rather than methodology. Option A is correct. Option B, kinesthetic learning, concerns hands-on engagement and is just one of several preferences; building a report around it would be unsuitable for most written or visual deliverables and would not address comprehension of the content. Option C, standardized templates, can promote consistency and efficiency, but a rigid template applied without regard to the audience may omit necessary context or include irrelevant detail. Effective practice is to identify the audience first, determine the decision it must make, and then select the language, depth, and visual form that serve that decision. Reference: Data Science Fundamentals Study Guide – Data Science Process: Communicating Results to Stakeholders and Tailoring Reports to the Audience.
===============
Where do the data input streams for data warehouses primarily come from?
Options:
third-party vendors.
web sites.
operational databases.
Answer:
CExplanation:
A data warehouse exists to integrate information about the business, and the most reliable and comprehensive record of business activity resides in the organization's operational systems. Transactional applications for sales, finance, human resources, supply chain, and customer service write their records to operational databases as events occur. Through scheduled or continuous ETL processes, those records are extracted, cleansed, conformed to common definitions, and loaded into the warehouse, where history is accumulated for analysis. Operational databases are therefore the primary feeder, and option C is correct. Third-party vendors, option A, supply valuable supplementary data such as demographic or market information, and web sites, option B, produce clickstream and campaign data that are often captured through analytics tools; both are legitimate secondary sources that enrich the warehouse, but neither provides the core transactional foundation. Modern architectures also ingest flat files, application programming interfaces, and event streams, which extends the range of inputs without displacing operational systems as the main source. The distinction matters for data quality, because any error in an operational record propagates to the warehouse, and so source system profiling and validation are standard early steps in warehouse design. Understanding the flow from operational systems to analytical systems explains why the two database types are separated. Reference: Data Science Fundamentals Study Guide – Data Management: Data Warehousing Concepts and Data Integration.
===============
Development for Power BI is restricted to which of the following operating systems?
Options:
iOS
Unix
Windows
Answer:
CExplanation:
Power BI is Microsoft's business analytics platform, and its components serve different roles. Power BI Desktop is the authoring application in which developers connect to data, shape it in Power Query, build data models, write DAX measures, and design reports. It is a native application that runs only on Windows, so report development is restricted to that operating system; users on macOS or Linux typically rely on a Windows virtual machine or a remote desktop session. Option C is correct. The Power BI service, delivered through a web browser, is used to publish, share, and collaborate on reports and allows limited editing from any operating system, and mobile applications for iOS and Android allow consumption and interaction with published content. These tools can therefore be used on other platforms, but they do not provide the full development environment. iOS, option A, is a mobile platform for which only the viewing application exists, and Unix, option B, is not supported at all. This constraint affects tooling decisions in organizations whose analysts use Apple or Linux machines, and it is a reason some teams choose cross-platform tools such as Tableau for authoring. Candidates should distinguish the desktop, service, and mobile components when answering questions about platform support. Reference: Data Science Fundamentals Study Guide – Data Management: Data Analysis and Business Intelligence Tools (Power BI).
===============
What is the PRIMARY purpose of statistical modeling?
Options:
describe the sampling techniques used.
convey analysis results and conclusions.
convey data correlation.
Answer:
BExplanation:
A statistical model is a mathematical representation of the relationships observed in data, and its principal function is to condense an analysis into a form that can be interpreted, evaluated, and communicated. A fitted regression equation, for instance, expresses how an outcome responds to changing predictors and gives stakeholders a defensible basis for conclusions, forecasts, and decisions. In that sense the model is the vehicle through which the outcomes of analysis are conveyed. Option A is incorrect because sampling is a design decision taken before modeling; documenting the technique supports credibility but is not why a model is built. Option C is too narrow. Correlation is one kind of relationship a model may express, yet models also capture causal structure under suitable designs, classify observations into categories, estimate probabilities, and generate predictions, none of which reduces to reporting correlation. Framing the purpose as conveying results and conclusions covers all of these uses and aligns with the study guide's emphasis on models as the bridge between raw analysis and actionable findings. Practitioners should also recall that a model communicates conclusions only to the extent that its assumptions and limitations are stated alongside its outputs. Reference: Data Science Fundamentals Study Guide – Data Science Concepts: Statistical Methods and Modeling.
===============
Statistical sampling refers to which of the following?
Options:
a data mining algorithm to cluster a given number of objects.
a further refinement of an earlier population sample.
a method of selecting a portion of a population.
Answer:
CExplanation:
Statistical sampling is the practice of drawing a subset of observations from a larger population so that the characteristics of the whole can be estimated without examining every member. The approach exists because a full census is frequently too costly, too slow, or physically impossible, whereas a carefully designed sample supports inference about population parameters such as the mean, proportion, or variance, together with a quantifiable margin of error. The reliability of those inferences depends on how the portion is chosen. Probability designs (simple random, systematic, stratified, cluster) give every member a known chance of selection, while non-probability designs (convenience, quota, judgmental) sacrifice rigor for speed. Option C states precisely this: selecting a portion of a population. Option A describes clustering, an unsupervised data mining technique that groups similar objects and plays no part in choosing records from a population. Option B mischaracterizes sampling as a second-pass narrowing of an existing sample; multistage designs exist, but refinement is not what defines the concept. Because the representativeness of the sample governs the validity of every later calculation, sampling is treated as a foundational step ahead of modeling and hypothesis testing. Reference: Data Science Fundamentals Study Guide – Data Science Concepts: Statistical Foundations, Populations and Sampling.
===============