Question 1

Business Intelligence (BI) data sources and boundaries Which of the following is NOT a way BI systems obtain or ingest data for analysis and reporting?

Accepted Answer

Introduction / Context: Business Intelligence (BI) focuses on turning data into information and insight. A key exam topic is distinguishing BI (analytics, reporting, dashboards, data mining) from OLTP systems (Online Transaction Processing), which capture and record day-to-day business events. This question tests whether you can separate data acquisition methods suitable for BI from operational transaction processing that belongs to OLTP. Given Data / Assumptions: BI systems consume data from various sources to analyze the past and present and sometimes to forecast the future. Operational systems (e.g., point-of-sale, order entry) exist to process and record transactions reliably and quickly. Organizations may also purchase external datasets to enrich analytics. Concept / Approach: BI systems generally read from operational databases, staging areas, or curated warehouses/lakes. They also use batch extracts and third-party data. However, they do not process transactions in the sense of writing order rows, debiting accounts, or updating inventory with ACID guarantees—those are OLTP responsibilities. Step-by-Step Solution: 1) Identify BI activities: reading, transforming, aggregating, modeling, and visualizing data.2) Identify OLTP activities: inserting and updating transactional records with strict latency and integrity requirements.3) Options A, C, D, and E all describe ways BI can obtain input datasets (direct reads, extracts, vendor data, lake files).4) Option B describes OLTP behavior—processing transactions—not BI ingestion. Verification / Alternative check: Enterprise architectures typically separate OLTP and BI/analytics stacks to avoid workload contention and to optimize each for its purpose. BI draws from OLTP but does not replace it. Why Other Options Are Wrong: Read operational DBs: common for reporting or ELT. Process extracts: standard ETL/ELT feed. Vendor data: widely used to enrich analysis. Lake/staging files: a normal analytics intake path. Common Pitfalls: Confusing reading from OLTP with processing transactions. Thinking BI must always copy data; some BI reads are virtual or federated. Final Answer: Process transactions (capture day-to-day orders, payments, inventory moves).

Question 2

Market Basket analysis terminology In 1,000 video rental transactions with items A, B, C, and D, the probability that Video D is rented given that Video C is rented (P(D | C)) is called:

Accepted Answer

Introduction / Context: Association rule mining (Market Basket analysis) uses three core measures—support, confidence, and lift—to quantify relationships like “if C then D.” Understanding the difference between these measures is essential for evaluating cross-sell opportunities and designing promotions. Given Data / Assumptions: A dataset of 1,000 transactions includes videos C and D among other items. We are interested in the conditional probability P(D | C). We assume standard definitions used in association rule mining. Concept / Approach: Support measures how often items occur together: support(C ∧ D) = count(C and D) / total_transactions. Confidence measures the rule strength when the antecedent occurs: confidence(C → D) = support(C ∧ D) / support(C) = P(D | C). Lift normalizes confidence by the base rate of D: lift(C → D) = confidence(C → D) / support(D). Step-by-Step Solution: 1) Translate the phrase “probability that D is rented given C is rented” into notation: P(D | C).2) Recall definitions: support = P(C ∧ D), confidence = P(D | C), lift = P(D | C) / P(D).3) Match the term to the definition: P(D | C) is confidence.4) Therefore, select “Confidence.” Verification / Alternative check: Any text on association rules (Apriori, FP-Growth) defines confidence as a conditional probability given the antecedent, aligning with P(D | C). Why Other Options Are Wrong: Basic probability: vague and not a standard metric here. Support: joint occurrence frequency, P(C ∧ D), not conditional. Lift: a ratio that compares confidence to the baseline frequency of D. Conviction: another quality metric, not P(D | C). Common Pitfalls: Confusing support with confidence because both use counts. Using high confidence rules with low support, leading to unstable insights. Final Answer: Confidence.

Question 3

Using operational data for BI The “curse of dimensionality” most closely relates to which problem when using operational data for analytics and reporting?

Accepted Answer

Introduction / Context: The “curse of dimensionality” refers to the challenges that arise as the number of features (dimensions) grows. In BI and data mining, high dimensionality can make distance metrics unreliable, models prone to overfitting, and queries computationally expensive. This question checks your conceptual mapping of that phrase to a typical analytics problem category. Given Data / Assumptions: Operational systems can generate many attributes per entity (customer, product, event). Analysts may be tempted to carry hundreds or thousands of features into BI models. We classify typical data quality/integration issues separately from dimensionality. Concept / Approach: As dimensionality increases, data becomes sparse; nearest-neighbor distances converge; and model generalization worsens without careful regularization, feature selection, and dimensionality reduction. Therefore, the phrase aligns most with “too much data” in the sense of too many columns/features, not merely volume of rows. Step-by-Step Solution: 1) Interpret “curse” as problems due to high number of features.2) Distinguish from dirty or inconsistent values (data quality issues).3) Distinguish from non-integration (system architecture issue).4) Conclude the best match is “Too much data (very high dimensional feature space).” Verification / Alternative check: Machine learning texts consistently equate the phrase with high-dimensional spaces that undermine intuitive geometry and require feature selection or dimensionality reduction (e.g., PCA). Why Other Options Are Wrong: Dirty/inconsistent data: quality problems, not dimensionality. Non-integrated data: siloing and schema mismatch, not feature count. Sparse keys: modeling concern but not the core of the “curse.” Common Pitfalls: Confusing “big data” (rows) with “high-dimensional data” (columns). Throwing all attributes into models without regularization. Final Answer: Too much data (very high dimensional feature space).

Question 4

Components of a data warehouse Which of the following is NOT a component of a data warehouse architecture?

Accepted Answer

Introduction / Context: A classic data warehouse ecosystem includes data ingestion and transformation tooling, the centralized analytical store, and rich metadata to document definitions, lineage, and usage. This question confirms that you recognize these standard parts and do not confuse them with OLTP application interfaces. Given Data / Assumptions: We consider a conventional warehouse (on-prem or cloud) with ETL/ELT pipelines. Metadata is critical for governance and discoverability. The core warehouse tables hold integrated, historical, conformed data. Concept / Approach: Options A, B, and C all describe canonical warehouse components. “None of the above” is therefore correct in the sense that none are invalid. Note that OLTP transaction screens (not listed among A–D) are not warehouse components, but that choice is not an answer option; it simply illustrates the contrast between analytics and transactions. Step-by-Step Solution: 1) Evaluate each listed component against standard warehouse architecture.2) ETL/ELT programs: yes.3) Warehouse data store: yes.4) Metadata: yes.5) Therefore select “None of the above; all are valid components.” Verification / Alternative check: Industry references and vendor architectures (e.g., Kimball/Inmon patterns, modern cloud warehouses) include ingestion, storage, and metadata management as core pieces. Why Other Options Are Wrong: Each of A–C is essential; rejecting any would break governance or pipelines. Common Pitfalls: Confusing ELT vs ETL processing patterns; both are within the ingestion component. Ignoring metadata and lineage, which harms trust and reuse. Final Answer: None of the above; all are valid data warehouse components.

Question 5

Data integration challenge You are asked to build a BI report, but required data lives partly in SQL Server and partly in Oracle. This situation is an example of:

Accepted Answer

Introduction / Context: BI work often begins with data integration—consolidating information spread across multiple operational systems. When the same subject area lives in separate platforms (e.g., SQL Server and Oracle), analytics teams must reconcile schemas, keys, and semantics before reporting. Given Data / Assumptions: Two RDBMS platforms: SQL Server and Oracle. Overlapping business entities and measures exist across them. Goal: unify data for a single report. Concept / Approach: This is a textbook case of non-integrated data. The problem is not necessarily quality (dirty) or encoding conflicts (inconsistent), though those may appear later; the primary challenge is that the data is siloed and must be integrated via ETL/ELT, virtualization, or federation. Step-by-Step Solution: 1) Identify that data resides in different systems.2) Recognize the need for integration (schema mapping, keys, conforming dimensions).3) Plan a pipeline or virtualization layer to deliver a unified model.4) Therefore, select “Non-integrated data.” Verification / Alternative check: Data warehousing methodologies emphasize integration as a central step (conformed dimensions, integrated fact tables) precisely because organizations frequently operate multiple OLTP systems. Why Other Options Are Wrong: Dirty/inconsistent: may occur but are secondary here. Wrong format: not the core issue—both are relational systems. Over-normalized: a modeling concern, not cross-system integration. Common Pitfalls: Attempting to report directly across silos without harmonization, leading to mismatched totals. Final Answer: Non-integrated data (data siloed across systems).

Question 6

Reading operational data difficulties Which of the following is NOT a typical reason that operational (OLTP) data is hard to read for BI purposes?

Accepted Answer

Introduction / Context: Operational databases are optimized for fast inserts/updates, not for analytics. BI teams commonly face quality issues, missing values, and data scattered across silos. This question asks you to spot the one choice that is not a problem. Given Data / Assumptions: Common OLTP problems include dirty data, nulls, and cross-system fragmentation. Analytics often needs denormalization and time-variant history. Duplicate records are a headache; non-duplicated data is actually good. Concept / Approach: “Non-duplicated data” describes a desirable state: having a single, correct record for each business entity. The other options describe real obstacles when preparing data for BI. Therefore, select the item that is clearly not a reason for difficulty. Step-by-Step Solution: 1) Compare each option against known OLTP pain points.2) Dirty data → problem.3) Missing values → problem.4) Non-integrated → problem.5) Wrong format/granularity → problem.6) Non-duplicated data → actually helpful, thus NOT a reason. Verification / Alternative check: Data quality frameworks aim to reduce duplicates, enforce constraints, and harmonize codes—improvements that make BI easier, not harder. Why Other Options Are Wrong: Dirty data: increases cleansing effort. Missing values: block accurate metrics. Non-integrated: requires heavy ETL. Wrong format: forces re-modeling and aggregation. Common Pitfalls: Assuming normalized OLTP schemas are directly consumable by BI without transformation. Final Answer: Non-duplicated data (no unnecessary repeats).

Question 7

RFM analysis purpose RFM analysis uses customers’ Recency, Frequency, and Monetary patterns primarily to:

Accepted Answer

Introduction / Context: RFM (Recency, Frequency, Monetary) is a lightweight but powerful technique for customer segmentation. By scoring how recently a customer purchased, how often they purchase, and how much they spend, marketers identify high-value groups for retention, cross-sell, and re-activation campaigns. Given Data / Assumptions: We have customer-level purchase histories. We can compute recency (time since last order), frequency (orders in a window), and monetary (spend in a window). We intend to prioritize outreach by likely value. Concept / Approach: RFM assigns each dimension a score (e.g., 1–5 buckets). Concatenated scores enable easy ranking: customers with high recency, high frequency, and high monetary are prime targets; those with low recency may need win-back campaigns. The method is descriptive ranking, not a specific predictive model like regression by itself. Step-by-Step Solution: 1) Compute R, F, M metrics over a defined time window.2) Bin each metric into score buckets (e.g., quintiles).3) Combine scores to produce segments (e.g., 5-5-5 elite, 1-1-1 at-risk).4) Rank customers by segment for targeted actions. Verification / Alternative check: CRM playbooks and analytics primers consistently present RFM as a segmentation and ranking tool used to improve campaign ROI, not as an OLTP or forecasting engine by itself. Why Other Options Are Wrong: Recording transactions: OLTP responsibility. Regression-only prediction: RFM is descriptive; prediction may use additional models. OLAP-only: RFM can be computed with or without cubes. Inventory reorders: separate operational analytics. Common Pitfalls: Using absolute thresholds instead of relative scoring, which can hide seasonal effects. Final Answer: Analyze and rank customers for segmentation and targeting.

Question 8

Scope of Business Intelligence (BI) Which of the following describes tasks that BI systems do NOT perform?

Accepted Answer

Introduction / Context: BI systems are designed for analysis—descriptive, diagnostic, and sometimes predictive or prescriptive. By contrast, OLTP systems are optimized to process and record transactions in real time with strict integrity and latency. This question checks if you can separate analytical workloads from transactional ones. Given Data / Assumptions: BI includes reporting, dashboards, ad-hoc queries, and data mining. OLTP includes order capture, payments, and inventory updates. “Only” in an option means that is the sole or primary focus. Concept / Approach: Option A (analyzing past/current) is a core BI activity. Option B (processing and recording transactions) is OLTP. Therefore, the task not performed by BI is B. Do not be tricked by “only” in A—BI indeed analyzes past and current activity (and may forecast), so A is still something BI does. Step-by-Step Solution: 1) Map each option to BI vs OLTP.2) A → BI; B → OLTP.3) The question asks what BI does not do → choose B.4) Options C/D are meta-answers; they do not fit once A and B are evaluated. Verification / Alternative check: Reference architectures keep analytical stores (warehouses/lakes) separate from operational systems to prevent workload interference and maintain ACID guarantees for transactions. Why Other Options Are Wrong: A: Clearly a BI function. C/D: Contradict the evaluation of A and B. E: Primary system of record is OLTP/ERP, not BI. Common Pitfalls: Thinking BI replaces OLTP; it complements it. Final Answer: Process and record transactions only (e.g., capture sales into the order table).

Question 9

RFM terminology In RFM analysis, the letter “F” stands for:

Accepted Answer

Introduction / Context: RFM stands for Recency, Frequency, and Monetary value. These three dimensions summarize customer purchase behavior in a compact way so that marketers can build effective segments without complex modeling. This question checks your recall of the basic acronym. Given Data / Assumptions: Standard RFM definitions apply. We interpret “F” as the number of orders in a given time window. We use RFM to compare customers on a common scale. Concept / Approach: “Frequency” counts how often a customer buys in the analysis window. Combined with “Recency” (how long since last order) and “Monetary” (total spend or average order value), it helps identify loyal, high-spend customers versus inactive or low-value ones. Step-by-Step Solution: 1) Recall the acronym: R = Recency, F = Frequency, M = Monetary.2) Map “F” to “Frequency,” not similar-sounding terms like “Freshness.”3) Select the correct option accordingly. Verification / Alternative check: Marketing analytics references and CRM tooling universally define “F” as Frequency, often operationalized as orders per period or count of purchase events. Why Other Options Are Wrong: Freshness: sometimes used informally for recency, not frequency. Fast food / Fantasy / Friction: unrelated distractors. Common Pitfalls: Mixing up “freshness” with “recency.” Final Answer: Frequency.

Question 10

Interpreting RFM scores After RFM scoring, a customer Mary Jones has a score of 1-1-5 (R-F-M). This implies Mary:

Accepted Answer

Introduction / Context: RFM scores typically bucket each dimension from best to worst (e.g., 1 = best, 5 = worst) or vice versa, depending on implementation. Many exam and textbook conventions use 1 as most desirable. Interpreting a composite like 1-1-5 correctly is a common test of understanding. Given Data / Assumptions: We adopt the common convention: 1 = best/most desirable, 5 = least desirable for each dimension. R = Recency, F = Frequency, M = Monetary value. Mary’s score is 1-1-5. Concept / Approach: Under the stated convention, R = 1 means very recent purchase; F = 1 means purchases are frequent; M = 5 means low spend (small orders) relative to peers. Therefore, Mary is active and loyal but with low order value—an ideal candidate for upsell to higher-value bundles or premium offers. Step-by-Step Solution: 1) Map R=1 → recent activity.2) Map F=1 → frequent buyer behavior.3) Map M=5 → low monetary value (small average or total spend).4) Combine into a plain-English interpretation: recent and frequent, but low order size → choose option C. Verification / Alternative check: Many RFM tutorials explicitly show “111” as the best and “555” as the weakest segment. Substituting 1-1-5 fits a high-engagement but low-spend profile. Why Other Options Are Wrong: A: Would require M=1 (high spend), not 5. B/D: Indicate poor recency/frequency, which contradicts R=1 and F=1. E: Contradiction—any positive frequency implies spend unless there are zero-value orders. Common Pitfalls: Reversing the scale (5 best). Always confirm the convention used. Final Answer: Has ordered recently, orders frequently, but does not place a large order when she orders.

Question 11

In Online Analytical Processing (OLAP), what are analysts primarily interested in when performing multidimensional analysis across cubes, hierarchies, and aggregation levels? Select the single best answer.

Accepted Answer

Introduction / Context: OLAP (Online Analytical Processing) supports interactive, multidimensional exploration of business data. Analysts slice, dice, drill down, and roll up to answer questions such as revenue by product, region, and time. Two foundational concepts drive OLAP: measures (numeric facts such as sales, cost, margin) and dimensions (business perspectives like time, product, geography, and customer) organized into hierarchies and levels. Given Data / Assumptions: Environment: OLAP cubes or tabular models. Measures: additive/semi-additive facts aggregated across dimensions. Dimensions: descriptive attributes with hierarchies (e.g., Year > Quarter > Month). User goal: meaningful analysis that combines numbers with context. Concept / Approach: OLAP analysis requires both the quantitative side (measures) and the qualitative context (dimensions). Measures deliver magnitude; dimensions provide the “by what/where/when/which/who.” Levels (e.g., month vs. day) are simply positions within a dimension hierarchy—useful, but not the standalone focus. Step-by-Step Solution: 1) Identify the quantitative facts: these are measures like Sales_Amount, Units_Sold, Profit.2) Identify the descriptive axes: these are dimensions such as Time, Product, Channel, Region.3) Realize that insights come from combining both: e.g., Profit (measure) by Region and Quarter (dimensions/levels).4) Conclude that OLAP concerns measures evaluated along dimensions (and their levels). Verification / Alternative check: Any OLAP report grid (pivot table) displays measures in cells and dimensions on rows/columns/filters. Remove dimensions and the numbers lose context; remove measures and there is nothing to aggregate. Why Other Options Are Wrong: Levels only: levels are part of dimensions, not the full scope. Dimensions only: context without numbers yields no analysis. Measures only: numbers without breakdown lack meaning. Common Pitfalls: Confusing “levels” with the entire analysis; thinking a single dimension (like Time) is sufficient for all insights; ignoring measure aggregation behavior (additive vs. semi-additive). Final Answer: measures and dimensions

Question 12

Within common data mining categories, which technique is considered unsupervised because it discovers structure in unlabeled data without target variables? Select the single best answer.

Accepted Answer

Introduction / Context: Data mining tasks are often divided into supervised (predictive) and unsupervised (descriptive) learning. Supervised methods learn from labeled examples to predict a target (e.g., churn yes/no or sales amount), while unsupervised methods discover natural groupings and patterns in data without predefined labels. Given Data / Assumptions: No explicit target variable for unsupervised learning. Candidate techniques: clustering, regression, and RFM analysis. Goal: identify the method that is inherently unsupervised. Concept / Approach: Cluster analysis (e.g., k-means, hierarchical clustering, DBSCAN) is unsupervised: it partitions records into groups based on similarity. Regression is typically supervised (linear, logistic, etc.), mapping features to a known numeric or categorical target. RFM is a business segmentation heuristic using Recency, Frequency, and Monetary ranks; it is descriptive rather than a canonical machine-learning algorithm and is not the standard example of unsupervised learning compared to clustering. Step-by-Step Solution: 1) Check whether a target label exists: unsupervised learning has none.2) Cluster analysis: groups unlabeled data → unsupervised.3) Regression: predicts a target (continuous or categorical via logistic) → supervised.4) RFM: descriptive ranking/segmentation; not a core ML algorithm; not the canonical unsupervised technique. Verification / Alternative check: Introductory ML texts list clustering and association rules as primary unsupervised methods; regression appears in supervised chapters. Why Other Options Are Wrong: Regression only: supervised by definition. RFM only: heuristic segmentation, not a standard unsupervised algorithm. Both Regression and RFM: mixes supervised with descriptive ranking; incorrect. Common Pitfalls: Assuming logistic regression is unsupervised because it outputs classes; it still requires labeled outcomes. Treating business heuristics (RFM) as ML categories. Final Answer: Cluster analysis only

Question 13

A video rental store has 1,000 transactions with items A, B, C, and D. The probability that C and D are rented together in the same transaction is called what in market-basket analysis? Select the single best answer.

Accepted Answer

Introduction / Context: Market-basket (association rule) analysis studies how items co-occur in transactions. Three core metrics are commonly used: support, confidence, and lift. Understanding these is crucial for interpreting association rules and prioritizing actionable insights in retail and recommendation systems. Given Data / Assumptions: Transactions: 1,000 total. Items: A, B, C, D. Event of interest: C and D in the same basket. Concept / Approach: Support measures how frequently an itemset appears in the dataset: support(C,D) = count(C ∩ D) / total_transactions. Confidence measures conditional probability for a rule: confidence(C → D) = support(C,D) / support(C). Lift evaluates the strength of a rule relative to independence: lift(C → D) = confidence(C → D) / support(D). Step-by-Step Solution: 1) Identify the question: probability that C and D occur together in the same transaction.2) Map to definition: joint probability of an itemset equals support of that itemset.3) Therefore, the correct term is support. Verification / Alternative check: Compute an example: if 120 of 1,000 baskets contain {C,D}, then support({C,D}) = 120 / 1,000 = 0.12 (12%). Confidence and lift need additional marginals and are not required here. Why Other Options Are Wrong: Basic probability: vague wording; the established term is support. Confidence: conditional on antecedent, not joint itemset frequency. Lift: ratio that adjusts confidence by expected co-occurrence under independence. Common Pitfalls: Confusing support with confidence; using high-confidence rules with very low support, which may be unreliable; ignoring lift when ranking rules for actionability. Final Answer: support

Question 14

Business Intelligence (BI) reporting analyses can be executed using what query capabilities in relational databases? Select the single best answer.

Accepted Answer

Introduction / Context: BI reporting transforms raw data into meaningful summaries and dashboards. Many required aggregations—totals, averages, groupings—are achievable with ANSI SQL. More advanced multi-dimensional summaries are often simplified by SQL extensions (e.g., GROUPING SETS, CUBE, ROLLUP). Given Data / Assumptions: Relational data warehouse or data mart. Need to produce standard and multi-level aggregations. Tooling may range from plain SQL editors to OLAP-aware engines. Concept / Approach: Standard SQL provides SELECT, GROUP BY, HAVING, and window functions for many reports. Extensions such as CUBE/ROLLUP/GROUPING SETS enable multi-hierarchy totals in a single query, reducing query proliferation and simplifying pivot-like outputs directly from the database. Step-by-Step Solution: 1) Recognize that ordinary reports (grouped totals) are expressible in standard SQL.2) Identify that multi-dimensional subtotaling is easier with SQL extensions.3) Conclude that both are valid and commonly used together in BI reporting. Verification / Alternative check: Examine typical BI SQL: a base GROUP BY for detail, plus ROLLUP for totals or CUBE for all combinations—both coexist in production systems. Why Other Options Are Wrong: Standard only: ignores powerful extensions. Extensions only: basic reports do not require them. OLAP only: BI reporting is broader than OLAP and often SQL-driven. Common Pitfalls: Duplicating queries for each subtotal instead of using GROUPING SETS; overlooking window functions for running totals, ranks, and percent-of-total calculations. Final Answer: both standard SQL and extensions to SQL

Question 15

A reporting system publishes reports on a website where users log in and retrieve them on demand. What report delivery mode is this? Select the single best answer.

Accepted Answer

Introduction / Context: Report delivery models determine how information reaches consumers. Two common patterns are “push” and “pull.” Understanding them helps design reporting workflows that fit user needs and system load. Given Data / Assumptions: Reports are hosted on a site. Users initiate access by browsing or clicking links. No automatic emailing or broadcasting is described. Concept / Approach: In a pull model, users request reports when needed (self-service portals, dashboards). In a push model, the system delivers reports proactively (scheduled emails, subscriptions, alerts). RFM and OLAP are analytics concepts, not delivery modes. Step-by-Step Solution: 1) Identify who initiates delivery: user or system.2) User-initiated access implies pull delivery.3) Therefore, the correct mode is Pull. Verification / Alternative check: Examples: logging into a portal or dashboard is pull; receiving a scheduled email report is push. Why Other Options Are Wrong: Push: system-initiated delivery. RFM: segmentation technique. OLAP: analytical approach, not a delivery mode. Common Pitfalls: Confusing “publish” with “push.” Publishing makes content available; it does not imply proactive distribution. Final Answer: Pull

Question 16

Which of the following operations is outside the scope of a Business Intelligence (BI) reporting system that reads from data warehouses and data marts? Select the single best answer.

Accepted Answer

Introduction / Context: BI reporting systems are primarily read-oriented: they aggregate, filter, and present data for decision-making. Distinguishing read-only operations from write operations helps maintain data integrity in analytical stores. Given Data / Assumptions: Source: data warehouse/mart optimized for reads. BI reports perform queries and aggregations. No transactional updates should occur from the reporting layer. Concept / Approach: Reporting tools run SELECT queries with filters, groups, and calculations. Data modification (INSERT/UPDATE/DELETE) belongs to ETL/ELT processes, data quality operations, or transactional systems—not to end-user reporting interfaces. Step-by-Step Solution: 1) Identify read operations: filtering, grouping, aggregating, sorting.2) Identify write operations: modifying persisted data.3) Match the reporting role: read-only, hence modifying data is out of scope. Verification / Alternative check: Enterprise BI platforms enforce read-only connections to warehouses in production; write access is restricted to ETL or admin roles. Why Other Options Are Wrong: Filter: core reporting function. Group: essential for aggregates. “Both filter and group” contradicts BI purpose. Common Pitfalls: Allowing ad-hoc users write permissions to analytics stores; mixing OLTP updates with analytical schemas (star/snowflake models). Final Answer: Modify data

Question 17

In the context of BI reporting and predictive analytics, which of the following is a popular classification technique that builds a model to predict a discrete outcome from features? Select the single best answer.

Accepted Answer

Introduction / Context: Classification assigns records to categories such as churn vs. non-churn, good vs. bad credit, or responder vs. non-responder. Many algorithms perform classification, including decision trees, naive Bayes, SVMs, and regression-based methods (notably logistic regression). Given Data / Assumptions: Goal: predict a categorical/boolean outcome. Candidate techniques: clustering, regression, RFM. Terminology uses “regression” broadly to include logistic regression for classification. Concept / Approach: Although “regression” often implies predicting a numeric value, logistic regression is a staple for binary classification. It models the log-odds of class membership from input features and outputs class probabilities. Cluster analysis, in contrast, is unsupervised segmentation without labeled targets; RFM is a descriptive ranking approach. Step-by-Step Solution: 1) Determine whether a labeled target exists (classification requires labels).2) Recognize logistic regression as a supervised classifier predicting class probabilities.3) Conclude that “regression analysis” (specifically logistic) is an appropriate classification technique. Verification / Alternative check: Industry practice: propensity models (e.g., likelihood to buy/churn) are commonly built with logistic regression due to interpretability and robustness. Why Other Options Are Wrong: Cluster analysis only: unsupervised; no class labels. RFM only: heuristic segmentation, not a predictive classifier. Both cluster and regression: mixing unsupervised with supervised; only regression fits classification among the options. Common Pitfalls: Assuming “regression” always means continuous targets; forgetting that evaluation for classification uses accuracy, AUC, precision/recall, not RMSE. Final Answer: regression analysis only

Question 18

Which tasks are commonly supported by data mining applications in customer analytics and decision support? Select the single best answer.

Accepted Answer

Introduction / Context: Data mining and advanced analytics systems go beyond simple transaction processing to derive insights and simulate outcomes. In marketing and CRM, two widely used activities are RFM-based segmentation and what-if (scenario) analysis for planning promotions, budgets, or pricing. Given Data / Assumptions: Use case: customer analytics. RFM: Recency, Frequency, Monetary scoring and segmentation. What-if: changing inputs/assumptions to estimate outcomes. Concept / Approach: While pure OLTP systems “process transactions,” analytics platforms apply algorithms and simulations to historical data to discover segments and forecast impacts. Many data mining suites integrate segmentation tools (including RFM-like ranking or clustering) and support what-if modeling through decision trees, regression, or integrated OLAP scenarios. Step-by-Step Solution: 1) Exclude options limited to transactional processing—those are OLTP responsibilities.2) Recognize that RFM segmentation and scenario analysis are both standard analytics tasks.3) Choose the option that includes both RFM and what-if analysis. Verification / Alternative check: Commercial analytics tools and open-source stacks (SQL + Python/R) routinely implement RFM scoring and scenario forecasting for marketing experiments. Why Other Options Are Wrong: Process transactions only: transactional, not data mining. RFM only: too narrow. What-if only: too narrow. Common Pitfalls: Confusing OLAP pivoting with predictive what-if modeling; treating RFM as a substitute for true clustering or predictive scoring when richer methods are needed. Final Answer: Do both RFM and what-if analysis

Question 19

Which of the following is not a capability or typical purpose of a BI reporting system in an enterprise analytics stack? Select the single best answer.

Accepted Answer

Introduction / Context: BI reporting systems transform data into dashboards and reports. They summarize, visualize, and distribute information rapidly. Predictive modeling, however, is a separate discipline typically handled by data mining or machine learning components. Given Data / Assumptions: BI reporting focuses on descriptive analytics. Predictive analytics requires statistical/ML algorithms and model training. Modern platforms may integrate both, but conceptually they are distinct layers. Concept / Approach: Reporting systems excel at aggregations, KPIs, and dissemination across devices and portals, including connecting to multiple (disparate) sources via semantic or data virtualization layers. Statistical prediction (e.g., forecasting, classification) belongs to data mining/ML pipelines, whose outputs may then be displayed by BI tools. Step-by-Step Solution: 1) Identify descriptive vs. predictive tasks.2) Map BI reporting to descriptive: create and deliver meaningful summaries, often across many sources.3) Recognize that predictive modeling is outside baseline reporting and thus is the “not” capability here. Verification / Alternative check: Architecture diagrams show ETL/ELT → warehouse → reporting for dashboards; separate ML services produce forecasts that reports can display but not train. Why Other Options Are Wrong: Create meaningful information: core purpose. Timely delivery: scheduling/alerting features support this. Use disparate sources: supported via connectors/semantic layers. Common Pitfalls: Assuming a dashboard tool that displays a forecast also built the model; conflating visualization with modeling. Final Answer: use statistical procedures to predict future events

Question 20

Reports from an enterprise reporting system are usually not delivered through which medium? Select the single best answer.

Accepted Answer

Introduction / Context: Enterprises distribute reports electronically for speed, cost, and security. Understanding the common delivery channels helps distinguish modern BI practices from legacy or impractical approaches. Given Data / Assumptions: BI environments emphasize electronic, near-real-time access. Channels include portals, dashboards, and email subscriptions. Physical courier delivery is atypical for dynamic reports. Concept / Approach: Web portals and digital dashboards enable pull access and real-time KPI monitoring. Email supports push delivery of scheduled PDFs/links. Courier services are slow, costly, and incompatible with frequently refreshed data—rarely used outside exceptional compliance packets or archival distributions. Step-by-Step Solution: 1) List standard BI delivery channels: portal, dashboard, email.2) Evaluate courier: physical transport not suited for timely analytics.3) Select “Commercial courier service” as the medium usually not used. Verification / Alternative check: Review typical BI platform features: SSO portals, role-based dashboards, email subscriptions, and APIs—no courier option. Why Other Options Are Wrong: Web portal: standard pull model. Digital dashboard: standard real-time display. E-Mail: standard push model. Common Pitfalls: Confusing print-and-ship of static financial reports with general BI; assuming paper distribution scales to daily/weekly refresh needs. Final Answer: Commercial courier service

Database Processing for BIS Questions