Back to Subject

Unit 3: Various Data Preprocessing Operations - Practice Quiz

ECAP792 60 Questions
0 Correct 0 Wrong 60 Left
0/60

1 What is the main purpose of data cleaning?

Data cleaning Easy
A. To permanently archive raw data
B. To display data in a chart
C. To correct or remove inaccurate data
D. To create additional data sources

2 Which issue is commonly handled during data cleaning?

Data cleaning Easy
A. File permissions
B. Screen resolution
C. Network speed
D. Missing values

3 What is a duplicate record?

Data cleaning Easy
A. A record stored in another format
B. A record containing only numbers
C. A record that appears more than once
D. A record arranged in sorted order

4 Which action can be used to handle a missing numerical value?

Data cleaning Easy
A. Convert it into a table name
B. Replace it with the column mean
C. Copy it into another database
D. Sort it in descending order

5 What does data integration involve?

Data integration Easy
A. Reducing the number of attributes
B. Deleting every incomplete record
C. Combining data from multiple sources
D. Dividing values into intervals

6 Which is an example of data integration?

Data integration Easy
A. Replacing missing ages with the mean
B. Merging customer tables from two branches
C. Converting temperatures from Celsius to Fahrenheit
D. Grouping salaries into income ranges

7 Which field is commonly used to match records across two tables?

Data integration Easy
A. A shared identifier
B. A random decimal
C. A display color
D. A file extension

8 What problem occurs when two sources use different names for the same attribute?

Data integration Easy
A. Schema inconsistency
B. Random sampling
C. Interval binning
D. Value normalization

9 What is data transformation?

Data transformation Easy
A. Protecting files with a password
B. Collecting data from new respondents
C. Converting data into a suitable form
D. Removing all records from a table

10 What does normalization usually do to numerical data?

Data transformation Easy
A. Changes numbers into missing values
B. Combines several database tables
C. Rescales values to a common range
D. Removes every repeated record

11 Which operation is an example of data transformation?

Data transformation Easy
A. Converting dates to one standard format
B. Selecting a sample from a population
C. Joining sales data with customer data
D. Deleting duplicate customer records

12 What is aggregation in data transformation?

Data transformation Easy
A. Summarizing detailed data into totals
B. Finding duplicate rows in a table
C. Splitting values into category intervals
D. Matching records from separate sources

13 What is the main goal of data reduction?

Data reduction Easy
A. To correct spelling mistakes in categorical values
B. To decrease data size while preserving useful information
C. To combine data stored in unrelated source systems
D. To increase data size by copying existing records

14 Which method reduces data by selecting only some records?

Data reduction Easy
A. Sampling
B. Discretization
C. Integration
D. Normalization

15 What does dimensionality reduction decrease?

Data reduction Easy
A. The number of databases
B. The number of data sources
C. The number of file types
D. The number of features

16 What is feature selection?

Data reduction Easy
A. Converting continuous values into categories
B. Combining records from different tables
C. Replacing missing values with estimates
D. Keeping the most relevant attributes

17 What does data discretization do?

Data discretization Easy
A. Combines tables from different databases
B. Converts continuous values into intervals
C. Corrects inaccurate values in records
D. Selects a subset of available features

18 Which is an example of discretizing age?

Data discretization Easy
A. Replacing missing ages with the median
B. Combining age data from two surveys
C. Removing age from the entire dataset
D. Grouping ages as child, adult, and senior

19 What is a bin in data discretization?

Data discretization Easy
A. An interval containing a range of values
B. A key connecting two data sources
C. A feature removed during sampling
D. A table containing duplicate records

20 Which type of data is most directly handled by discretization?

Data discretization Easy
A. Continuous numerical data
B. Encrypted password data
C. Unstructured image data
D. Relational schema data

21 A customer-income column contains 8% missing values and is strongly right-skewed because of a few very high incomes. Which imputation method is most appropriate?

Data cleaning Medium
A. Replace missing values with the median
B. Replace missing values with zero
C. Replace missing values with the mean
D. Remove every row containing a missing value

22 A dataset contains two records with the same customer ID, email address, and date of birth, but slightly different spellings of the customer name. What is the best cleaning action?

Data cleaning Medium
A. Replace both names with missing values
B. Merge them after verifying they represent one customer
C. Treat both records as unrelated customers
D. Delete both records from the dataset

23 A hospital dataset contains patient ages of 34, 67, 12, and 240 years. Which rule would most directly identify the invalid value?

Data cleaning Medium
A. A frequency constraint on age
B. A valid-range constraint on age
C. A formatting constraint on age
D. A uniqueness constraint on age

24 A temperature sensor occasionally records 999 when it loses connection. The normal operating range is to degrees Celsius. How should 999 initially be handled?

Data cleaning Medium
A. Scale it using min-max normalization
B. Keep it as a genuine outlier
C. Convert it to a missing value
D. Replace it with the maximum value

25 One sales table uses the field customer_id, while another uses client_number for the same identifier. What should be done before integrating the tables?

Data integration Medium
A. Normalize both fields to the interval
B. Remove both fields before joining
C. Discretize both fields into equal-width bins
D. Standardize the fields using schema matching

26 Two regional datasets store sales in dollars and euros. What is required before calculating combined total sales?

Data integration Medium
A. Convert both columns to a common currency
B. Join both datasets without changing values
C. Replace both columns with category labels
D. Sort both columns in ascending order

27 A customer address differs between a CRM record updated yesterday and a billing record updated six months ago. If update timestamps are reliable, which conflict-resolution rule is most suitable?

Data integration Medium
A. Select either address at random
B. Prefer the alphabetically first address
C. Keep only the shortest address
D. Prefer the most recently updated record

28 Joining an orders table to a promotions table unexpectedly multiplies each order into several rows because multiple promotion records share the same code. What should be checked first?

Data integration Medium
A. Whether numerical columns have been normalized
B. Whether the join key is unique as expected
C. Whether missing values use the median
D. Whether categories have been one-hot encoded

29 A model uses height in centimeters and annual income in dollars. Which transformation gives both variables mean and standard deviation ?

Data transformation Medium
A. Z-score standardization
B. Decimal scaling
C. Min-max normalization
D. Equal-width discretization

30 A transaction-amount variable is positive and strongly right-skewed. Which transformation is most likely to reduce the skew while preserving value order?

Data transformation Medium
A. A label-encoding transformation
B. A random permutation
C. A one-hot transformation
D. A logarithmic transformation

31 A feature named transport_type has the unordered values bus, train, car, and bicycle. Which encoding avoids implying an artificial ranking?

Data transformation Medium
A. Rank transformation
B. One-hot encoding
C. Ordinal encoding
D. Cumulative encoding

32 Using min-max normalization to , what is the transformed value of when the observed minimum is and maximum is ?

Data transformation Medium
A.
B.
C.
D.

33 A dataset has 100 highly correlated numerical features. You want fewer features while retaining at least 95% of the total variance. Which method is most appropriate?

Data reduction Medium
A. Equal-width binning
B. Principal component analysis
C. Mean imputation
D. One-hot encoding

34 A fraud dataset contains only 1% fraudulent transactions. You need a smaller representative sample for exploration. Which sampling method best preserves the class distribution?

Data reduction Medium
A. Convenience sampling
B. Systematic majority-only sampling
C. Simple sequential sampling
D. Stratified random sampling

35 A retailer stores one row per transaction but needs only monthly revenue trends by store. Which reduction operation is most suitable?

Data reduction Medium
A. Aggregate transactions by store and month
B. Encode each transaction using one-hot vectors
C. Duplicate transactions with missing store values
D. Normalize each transaction to unit length

36 A classification dataset includes a unique row ID, several useful measurements, and two redundant copies of existing features. Which action best reduces dimensionality without losing useful information?

Data reduction Medium
A. Remove the ID and redundant features
B. Duplicate the categorical features
C. Remove all continuous measurements
D. Keep only the unique row ID

37 Values range from through . Using five equal-width intervals, which interval contains the value ?

Data discretization Medium
A.
B.
C.
D.

38 A highly skewed income variable must be divided into four bins containing approximately the same number of records. Which method should be used?

Data discretization Medium
A. Equal-frequency discretization
B. Fixed-threshold discretization
C. Equal-width discretization
D. Random-boundary discretization

39 You are discretizing age for a churn classifier and want bin boundaries that best separate churned from non-churned customers. Which approach is most suitable?

Data discretization Medium
A. Unsupervised equal-width discretization
B. Alphabetical category ordering
C. Random interval discretization
D. Supervised entropy-based discretization

40 A discretization scheme uses the intervals , , and . Into which bin should the value be placed?

Data discretization Medium
A. The interval
B. The interval
C. Both adjacent intervals
D. The interval

41 A dataset has values missing at random, where the probability that feature is missing depends on observed feature . During 5-fold cross-validation, which procedure best prevents information leakage while preserving this relationship?

Data cleaning Hard
A. Fit a multivariate imputer on each training fold and apply it to that fold's validation data
B. Fit a multivariate imputer once on all predictors after removing the target variable
C. Remove incomplete rows separately from both the training and validation portions of each fold
D. Fit a median imputer on the complete dataset before creating the cross-validation folds

42 A transaction table contains two rows with the same customer, timestamp, and amount but different transaction IDs. Some legitimate customers can make identical purchases within the same second. What is the safest cleaning action?

Data cleaning Hard
A. Retain both rows until transaction-ID provenance or source-system rules establish duplication
B. Average the amounts and replace both records with one consolidated transaction
C. Delete one row because the three business attributes form a reliable composite key
D. Keep the record with the larger transaction ID and discard the other record

43 A pressure sensor occasionally reports to encode failure, while valid measurements are approximately Gaussian with mean and standard deviation . Which rule most appropriately handles the value ?

Data cleaning Hard
A. Replace it with the sample median because it is an extreme statistical outlier
B. Winsorize it to the lower boundary defined by three standard deviations
C. Convert it to missing using the sensor's documented sentinel-value rule
D. Retain it because automatic modification would alter the empirical distribution

44 After Unicode normalization and case folding, the customer labels MASS and Maß both map to the same normalized token. The distinction may be meaningful in the source system. What should the cleaning pipeline do?

Data cleaning Hard
A. Flag the collision and preserve original values while resolving it with domain rules
B. Select the most frequent spelling and replace every colliding source value
C. Merge the records because normalized equality guarantees semantic equivalence
D. Discard both records because their canonical representations are not unique

45 Table has three rows for customer C7, and table has four rows for C7. An inner join is performed using only customer_id. How many joined rows for C7 result, and what does this indicate?

Data integration Hard
A. rows, indicating an append operation at the customer grain
B. rows, indicating that table determines the output grain
C. rows, indicating a many-to-many join at the selected grain
D. rows, indicating that table determines the output grain

46 Two hospitals store temperature under a column named temp: one records Celsius and the other Fahrenheit. Both columns are numeric and have no missing values. Which integration step is essential before concatenation?

Data integration Hard
A. Resolve the semantic unit mismatch and convert both columns to a common unit
B. Rename both columns to a shared canonical name without changing values
C. Standardize each column independently to zero mean and unit variance
D. Cast both columns to the same floating-point representation before appending

47 For each loan application at time , a credit-score history table contains multiple timestamped scores. Which join avoids future-data leakage while using the freshest available score?

Data integration Hard
A. Join the score with the greatest timestamp satisfying
B. Join the score with the greatest timestamp satisfying
C. Join the maximum score recorded for that applicant across all timestamps
D. Join the score with the smallest absolute timestamp difference from

48 In entity resolution, pairwise rules classify A–B and B–C as matches but A–C as a non-match. Why can taking connected components over matched pairs be unsafe?

Data integration Hard
A. Connected components require every record pair to have identical attribute values
B. A component algorithm necessarily removes records having incomplete identifiers
C. Match decisions need not be transitive, so the component may merge distinct entities
D. Pairwise similarities cannot be represented by an undirected graph structure

49 A Box–Cox transformation is selected for a positive predictor during model development. Which deployment procedure is statistically valid?

Data transformation Hard
A. Estimate on training data and reuse that value for validation and future data
B. Set for validation data whenever its mean differs from the training mean
C. Estimate independently for every validation batch to match its distribution
D. Estimate on the combined training and validation data before evaluation

50 A categorical variable with levels is one-hot encoded into all indicator columns and used in an unregularized linear regression that includes an intercept. What is the resulting issue?

Data transformation Hard
A. The residual variance becomes zero because every row activates one indicator
B. The design matrix is rank-deficient because the indicators sum to the intercept
C. The model becomes nonlinear because each category receives a separate coefficient
D. The feature matrix loses sparsity because the intercept duplicates every category

51 A min–max scaler is fit to training values ranging from to . A valid deployment observation has value . Using , how should it be transformed if clipping was not part of the specification?

Data transformation Hard
A. Refit the maximum to and transform the observation to
B. Transform it to using the stored training minimum and maximum
C. Mark it missing because values outside the training range are undefined
D. Transform it to because every min–max output must lie in

52 A model is trained by ordinary least squares on , with residuals approximately . To estimate the conditional arithmetic mean of under this model, which back-transformation is appropriate?

Data transformation Hard
A.
B.
C.
D.

53 A low-variance feature almost perfectly separates a rare positive class, while several high-variance noise features dominate the dataset. Why can retaining only the leading principal components harm classification?

Data reduction Hard
A. PCA optimizes class separation but becomes unstable when classes are imbalanced
B. PCA preserves only nonlinear relationships and removes linear class information
C. PCA maximizes predictor variance without considering whether directions predict the target
D. PCA necessarily deletes every feature whose marginal variance is below one

54 A very large term-document matrix is sparse. The goal is a latent semantic representation without materializing a dense centered matrix. Which method is most suitable?

Data reduction Hard
A. Apply truncated SVD directly to the uncentered sparse matrix
B. Compute pairwise document covariance and retain all nonzero eigenvectors
C. Center every term column and apply a full covariance eigendecomposition
D. Convert the matrix to dense form and apply classical multidimensional scaling

55 Two disjoint groups summarize a numeric variable using and , where . Which additional information is required to compute the exact combined sample variance?

Data reduction Hard
A. Only the minimum and maximum from each group are additionally required
B. No additional information is required because the two summaries are sufficient
C. Every original observation is required because variance summaries cannot be merged
D. Only the median and interquartile range from each group are additionally required

56 A population contains positive cases. A reduced training sample includes all positives but only of negatives. To estimate population-level risk with an unweighted loss replaced by a weighted loss, which relative weighting is appropriate?

Data reduction Hard
A. Give each sampled negative ten times the weight of each sampled positive
B. Give both classes equal weight because all positive cases were retained
C. Give each sampled positive ten times the weight of each sampled negative
D. Give each negative one hundred times the weight of each sampled positive

57 A feature has values . An algorithm requests four equal-frequency bins while prohibiting identical values from being split across bins. What is the key consequence?

Data discretization Hard
A. The tied values must be distributed uniformly among bins to preserve quantile boundaries
B. Equal-width binning becomes identical to equal-frequency binning for this sample
C. Four equal-frequency bins always exist because the sample size is divisible by four
D. Exactly equal bin counts may be impossible because the four tied values must remain together

58 A supervised discretizer chooses cut points by maximizing information gain with respect to the target. Where must cut-point learning occur during cross-validation?

Data discretization Hard
A. Once on the full dataset, followed by reusing the cut points in every fold
B. Once on target-stratified predictors after labels have been removed from the dataset
C. Within each validation fold, followed by applying the cut points to its training fold
D. Within each training fold, followed by applying those cut points to its validation fold

59 A candidate split partitions a node containing 50 positives and 50 negatives into children and . Using base-2 entropy, what is the information gain?

Data discretization Hard
A. Approximately bits
B. Approximately bits
C. Exactly bit
D. Exactly bits

60 Training values are discretized using boundaries , , and . Which assignment correctly handles a deployment value of without refitting?

Data discretization Hard
A. Create a new bin based on the deployment value
B. Treat as missing because it exceeds the observed training maximum
C. Assign to using the stored boundaries
D. Rescale all boundaries so that becomes the upper endpoint
Previous MCQ
Unit 2
Next MCQ
Unit 4