Font size
WorksheetsMSTIP - FinalExam - Data Mining
Total questions: 100
Worksheet time: 2hrs 40mins
Which of the following is NOT a typical application of data mining?
Fraud detection
Weather forecasting
Market basket analysis
Customer segmentation
What is the main difference between data mining and traditional data analysis?
Data mining is faster
Data mining uses statistical methods only
Data mining discovers hidden patterns automatically
Data mining replaces databases
Which of the following is a descriptive data mining task?
Classification
Regression
Clustering
Prediction
Which term refers to the process of cleaning and transforming raw data before mining?
Data mining
Data integration
Data pre-processing
Data warehousing
In classification, the output variable is:
Continuous
Categorical
Text
Unsupervised
Which of the following is an example of a supervised learning task in data mining?
Clustering
Association rule mining
Classification
Dimensionality reduction
What is the purpose of association rule mining?
Predicting a numeric value
Discovering relationships among variables in large databases
Segmenting similar customers
Reducing data dimensions
The term “support” in association rule mining refers to:
The number of items in a transaction
The number of rules generated
The frequency of occurrence of an itemset
The accuracy of classification
Which technique is best suited for detecting unusual patterns that do not conform to expected behavior?
Clustering
Classification
Outlier detection
Association rules
Which data mining technique involves grouping similar data points without predefined labels?
Classification
Clustering
Regression
Decision tree
A confusion matrix is used to evaluate the performance of:
Clustering
Classification models
Association rule mining
Regression analysis
What does “overfitting” refer to in data mining?
The model fits training data well but performs poorly on new data
The model does not fit the training data well
The model predicts only categorical values
The data is too large to process
Which of the following is a characteristic of a good data mining algorithm?
It should work only with numeric data
It should overfit the data
It should generalize well to new data
It should ignore outliers
Which of these is a common algorithm for classification?
Apriori
K-Means
Decision Trees
PCA
The K in the K-Means algorithm represents:
Number of rules
Number of neighbors
Number of clusters
Number of attributes
Which data mining task is best for predicting a numerical outcome?
Classification
Regression
Clustering
Association
In data warehousing, the process of collecting and managing data from varied sources is called:
Data mining
Data integration
Data pruning
Data visualization
Principal Component Analysis (PCA) is primarily used for:
Classification
Regression
Dimensionality reduction
Clustering
Which one of the following is NOT a typical step in the data mining process?
Data collection
Data preprocessing
Modeling
Software debugging
What does the term “curse of dimensionality” refer to?
Too little data to analyze
Data with too many missing values
Increased data volume reducing model performance
Increased number of attributes causing performance issues
Which of the following is an example of a classification problem in supervised learning?
Predicting house prices
Forecasting stock values
Identifying spam emails
Estimating rainfall amounts
In supervised learning, the output variable is:
Unlabeled
Always categorical
Labeled
Always continuous
Which metric is most appropriate for evaluating a regression model?
Accuracy
F1 Score
Mean Absolute Error (MAE)
Precision
Which algorithm is primarily used for classification tasks?
Linear Regression
Logistic Regression
K-Means Clustering
Principal Component Analysis
Overfitting in a model refers to:
High training error, low test error
Low training error, high test error
Equal error in training and test sets
Model ignoring noise in training data
What is the purpose of cross-validation?
To reduce dimensionality
To improve training speed
To evaluate model performance on unseen data
To normalize data
Which of the following is not a supervised learning algorithm?
Decision Tree
Naive Bayes
Support Vector Machine
Apriori
In a confusion matrix, a false positive means:
The model predicted negative, but it was positive
The model predicted positive, but it was negative
Both prediction and actual were negative
Both prediction and actual were positive
What does the "bias-variance trade-off" describe?
Trade-off between training and test data
Trade-off between number of features and samples
Trade-off between underfitting and overfitting
Trade-off between classification and regression
Which supervised learning algorithm is most sensitive to feature scaling?
Decision Trees
K-Nearest Neighbors
Naive Bayes
Random Forest
Which of the following is a disadvantage of k-NN?
It is parametric
Training is very slow
It doesn't support multi-class classification
It cannot be used for regression
The ROC curve is useful for evaluating:
Regression models
Clustering algorithms
Binary classifiers
Dimensionality reduction
Which of the following helps prevent overfitting in decision trees?
Increasing depth
Increasing features
Pruning
Removing cross-validation
Which method combines predictions from multiple models to improve accuracy?
Feature selection
Ensemble learning
Regularization
Cross-validation
Which algorithm is a type of ensemble method?
Naive Bayes
KNN
Random Forest
Logistic Regression
The learning rate in gradient descent:
Controls how much we adjust weights with respect to the loss
Determines the model's accuracy
Specifies the number of features
Is fixed in all models
L1 regularization encourages:
Grouping of correlated features
Smooth decision boundaries
Sparse feature selection
Equal feature importance
Which of the following can help improve the generalization of a model?
Using the training set for evaluation
Increasing model complexity
Data augmentation
Removing regularization
Which loss function is commonly used in logistic regression?
Mean Squared Error
Hinge Loss
Cross-Entropy Loss
L1 Loss
What is the main objective of supervised learning?
Group similar items
Explore data without labels
Predict output values from labeled inputs
Reduce dimensionality
Which of the following is a typical use case of unsupervised learning?
Predicting stock prices
Classifying spam emails
Customer segmentation
House price prediction
In clustering, what is the primary goal?
To predict future outcomes
To group similar instances together
To reduce dimensionality
To classify data using labeled examples
Which of the following algorithms is commonly used for clustering?
K-Means
Linear Regression
Naive Bayes
Decision Tree
In Principal Component Analysis (PCA), what is the principal component?
The first variable in the dataset
A linear combination of original features
A decision boundary
A cluster center
What metric does the K-Means algorithm minimize?
Mean Absolute Error
Euclidean distance from centroids
Entropy
Gini index
Which of the following is a method for dimensionality reduction?
K-Nearest Neighbors
PCA
Random Forest
Logistic Regression
The elbow method is used to determine:
Feature importance
Optimal number of clusters in K-Means
Outliers
Classification accuracy
Hierarchical clustering differs from K-Means in that:
It requires predefined number of clusters
It only works with supervised data
It creates a dendrogram
It is faster for large datasets
Which of the following statements is true about DBSCAN?
It assumes spherical clusters
It requires specifying the number of clusters
It is sensitive to noise
It can find clusters of arbitrary shapes
What is the role of the 'epsilon' parameter in DBSCAN?
Defines the number of clusters
Specifies the minimum variance
Defines the neighborhood radius
Determines the step size
Which of the following is not typically an unsupervised learning technique?
K-Means
PCA
DBSCAN
Linear Discriminant Analysis (LDA)
PCA transforms the data to:
A lower-dimensional space with dependent features
A higher-dimensional space
A lower-dimensional space with uncorrelated features
The original feature space with normalized values
A key disadvantage of K-Means is:
It can find non-linear boundaries
It is deterministic
It requires labeled data
It assumes clusters are spherical and equal in size
Which evaluation metric is used for clustering when labels are unavailable?
Accuracy
Adjusted Rand Index
Silhouette Score
ROC-AUC
Which algorithm is best suited for discovering clusters with varying density?
K-Means
PCA
DBSCAN
Hierarchical clustering
Which of the following is true for hierarchical clustering?
It cannot be visualized
It doesn’t require a distance metric
It always produces the same number of clusters
It can be agglomerative or divisive
The curse of dimensionality affects:
Linear regression performance
K-Means efficiency and distance accuracy
Label encoding in supervised models
Ensemble model performance
What does t-SNE do?
Clusters data points
Classifies data
Projects high-dimensional data into 2 or 3 dimensions for visualization
Removes noise from data
In PCA, what determines how many components should be selected?
Number of classes
Explained variance ratio
Number of original features
Training error
Which technique can detect anomalies in data using unsupervised learning?
K-Nearest Neighbors
DBSCAN
Logistic Regression
Random Forest
In association rule mining, what does the "support" of a rule indicate?
The strength of the rule
The proportion of transactions that contain the itemset
The frequency of the consequent
The confidence level of the rule
Which of the following measures is used to evaluate the reliability of a rule?
Support
Confidence
Lift
All of the above
If an itemset has high support but low confidence, it means:
The rule is strong
The rule is rare
The rule occurs frequently but not reliably
The rule is invalid
What does "lift" measure in association rules?
Probability of consequent given antecedent
Ratio of observed support to expected support under independence
Support of the antecedent
Confidence of the rule
In the Apriori algorithm, what property helps reduce the number of candidate itemsets?
Downward closure property
Monotonicity
Lift maximization
Confidence boosting
Which algorithm improves upon Apriori by reducing the number of database scans?
FP-Growth
Eclat
A-PrioriTid
Naïve Bayes
What is a key difference between Apriori and FP-Growth?
Apriori uses a tree structure
FP-Growth uses candidate generation
Apriori is faster on sparse data
FP-Growth avoids candidate generation
Which of the following is not an association rule evaluation metric?
Conviction
Cosine
Entropy
Lift
What is the "confidence" of the rule A → B if support(A ∪ B) = 0.2 and support(A) = 0.4?
0.1
0.5
0.8
0.6
The FP-tree structure is used to:
Compress the transaction database
Replace itemsets with item graphs
Increase the number of candidate itemsets
Sort rules in descending lift
In association rule mining, what is a “frequent itemset”?
An itemset that occurs less than 50% of the time
An itemset with a support above a user-defined threshold
A set with high confidence
A rule with low lift
Which of the following best describes the “downward closure” property?
All subsets of a frequent itemset must also be frequent
All supersets of an infrequent itemset must be infrequent
Both A and B
None of the above
In the rule A → B, which itemset is the antecedent?
A
B
A ∪ B
B ∪ C
The rule A → B is more interesting if:
Lift ≈ 1
Confidence ≈ 0
Support is very low
Lift > 1
Which approach does FP-Growth use to mine frequent patterns?
Bottom-up search
Top-down search
Divide-and-conquer
Breadth-first search
What is the main disadvantage of the Apriori algorithm?
Requires sorting
Generates too many candidate itemsets
Cannot find frequent patterns
Works only for binary data
Conviction is used in association rule mining to measure:
Support strength
Rule reliability if consequent is false
Co-occurrence
Lift ratio
Which of the following is an example of an association rule?
{milk, bread}
milk → bread
Support({milk, bread}) = 0.4
{milk, butter, eggs}
What type of data is typically used in market basket analysis?
Sequential
Time-series
Transactional
Graph-based
Which of the following would indicate an interesting rule?
Low support, low confidence, lift = 1
High support, low lift
High confidence, lift > 1
Confidence = support
Which metric is best used when the classes are imbalanced?
Accuracy
Precision
Recall
F1 Score
What does a high precision score indicate?
Few false positives
Few false negatives
High true positives
Low accuracy
Which metric is calculated as: TP / (TP + FN)?
Precision
Recall
Specificity
Accuracy
What is the formula for accuracy?
TP / (TP + FN)
TP / (TP + FP)
(TP + TN) / (TP + TN + FP + FN)
TN / (TN + FP)
Which of the following is not an evaluation metric?
Confusion Matrix
Precision
Gradient
ROC AUC
What does ROC curve plot?
Precision vs Recall
True Positive Rate vs False Positive Rate
Accuracy vs Loss
Recall vs Specificity
Which metric is used to evaluate the ranking ability of a classifier?
F1 Score
ROC AUC
Accuracy
Precision
What does a confusion matrix help you visualize?
Model complexity
Prediction errors
Training time
Number of epochs
If a model has high recall but low precision, what does it mean?
Many relevant items are missed
Many irrelevant items are classified as relevant
High number of true negatives
Perfect accuracy
Which metric combines both precision and recall?
Accuracy
Specificity
F1 Score
ROC AUC
Specificity is defined as:
TN / (TN + FP)
TP / (TP + FN)
TP / (TP + FP)
FP / (FP + TN)
Which metric is best for minimizing false negatives?
Precision
Specificity
Recall
F1 Score
In a confusion matrix, which value represents false positives?
Top-left
Top-right
Bottom-left
Bottom-right
What does an AUC score of 0.5 indicate?
Excellent model
Random guessing
Perfect recall
High accuracy
Which metric increases if you increase the threshold in a classifier?
Recall
False Positive Rate
Precision
F1 Score
What happens to recall when the decision threshold is lowered?
Increases
Decreases
Remains same
Cannot be determined
Which scenario best suits using the Precision metric?
Detecting cancer cases
Spam email detection
Face recognition with many classes
Weather forecasting
What is the primary weakness of accuracy in imbalanced datasets?
Too complex to calculate
Ignores true positives
Doesn’t distinguish between classes
Misleading when one class dominates
What is the ideal value of the F1 Score?
0
0.5
1
Depends on model
Which metric would you use if false positives are costlier than false negatives?
Recall
Accuracy
Precision
F1 Score
