Test
Sin título
Crea una cuenta gratis para terminar este test
Guarda tu progreso y consulta tu resultado final.
¿Ya tienes cuenta? Inicia sesión
Preguntas del test (68)
1If you apply the ZeroR algorithm to a two-class data set and the error on the training set turns out to be 15%, it means that:
(select the correct answer!)- AWe got a pretty good model of the data.
- BThe data set is unbalanced.
- CIt means nothing, because we should
have test it on some independent test set. - DThe classification accuracy of ZeroR is 80%.
2When we try to model the data by using a supervised learning algorithm ...
(select the incorrect answer!)- A... the class vales for the training set examples are usually known.
- B... the dependent variable (class) has to be a numeric attribute.
- C... we try to learn a classifier that will "do well" on new "unseen" examples.
- D... we typically evaluate the built model on "unseen" -- test data.
3For a perfectly balanced two-class data set ZeroR will always output an error of:
- A33.33%
- B0%
- C50%
- D100%
4What does it mean "the probability of class-value 'YES' is 50%"?
- AThat half of all the examples in such a data set have the class-value 'YES'.
- BThat 50% of all the examples in such a data set are "positives".
- CThat ZeroR will make an error of 50% on such a data set.
- DThat the data set in question is a two-class data set.
5We ran the OneR algorithm on a data set with 4 attributes (A1, A2, A3 and A4) and the class C, and got the following errors:
A1 ... 50%,
A2 ... 10%,
A3 ... 10%,
A4 ... 25%.
Which attribute will OneR choose as the "best" attribute?- Aa random choice between A2 and A3
- BA3
- CA2
6How does the algorithm OneR deal with missing values?
(select the correct answer!)
Select one:- Ait treats them as a separate value of an attribute
- Bit tries to substitute them with an actual attribute value
- Cit ignores every example that contains at least one missing value
- Dit ignores them
7How does the OneR algorithm "process" numeric attributes?
(select the correct answer!)- Ait discretizes them using equal frequency discretization
- Binstead of calculating the error (%), it uses the correlation coefficient
- Cit discretizes them using class-dependent discretization
- Dit ignores them
8Identify the incorrect statement about the Naïve Bayes algorithm.
- ANaïve Bayes uses just some (the "best") attributes to predict the class.
- BNaïve Bayes uses all the attributes to predict the class.
- CNaïve Bayes assumes that all attributes are "equally important".
- DNaïve Bayes is (theoretically) based on probability theory.
9Information gain is ...
- A.. the "best" measure to select attributes in the process of building a decision tree.
- B... the information gained when classifying "new" examples with a decision tree.
- C... measured in bits.
- D... a measure for evaluating the performance of a decision tree.
10The entropy of having 4 equally likely events is:
- A0
- B2
- C1
- D4
11When we try to model the data by using a supervised learning algorithm ...
(select the incorrect answer!)- A... the class vales for the training set examples are usually known.
- B... we typically evaluate the built model on "unseen" -- test data.
- C... we try to learn a classifier that will "do well" on new "unseen" examples.
- D... the dependent variable (class) has to be a numeric attribute.
12For a perfectly balanced two-class data set ZeroR will always output an error of:
(select the correct answer!)- A33.33%
- B50%
- C0%
- D100%
13What does it mean "the probability of class-value 'YES' is 50%"?
(select the correct answer!)- AThat ZeroR will make an error of 50% on such a data set.
- BThat 50% of all the examples in such a data set are "positives".
- CThat half of all the examples in such a data set have the class-value 'YES'.
- DThat the data set in question is a two-class data set.
14We ran the OneR algorithm on a data set with 4 attributes (A1, A2, A3 and A4) and the class C, and got the following errors:
A1 ... 50%,
A2 ... 10%,
A3 ... 10%,
A4 ... 25%.
Which attribute will OneR choose as the "best" attribute?
(select the correct answer!)
Select one:- AA3
- Ba random choice between A2 and A3
- CA2
- DA1
15How does the algorithm OneR deal with missing values?
(select the correct answer!)- Ait ignores every example that contains at least one missing value
- Bit ignores them
- Cit tries to substitute them with an actual attribute value
- Dit treats them as a separate value of an attribute
16How does the OneR algorithm "process" numeric attributes?
(select the correct answer!)- Ait discretizes them using equal frequency discretization
- Binstead of calculating the error (%), it uses the correlation coefficient
- Cit ignores them
- Dit discretizes them using class-dependent discretization
17Identify the incorrect statement about the Naïve Bayes algorithm.
Select one:- ANaïve Bayes uses all the attributes to predict the class.
- BNaïve Bayes uses just some (the "best") attributes to predict the class.
- CNaïve Bayes assumes that all attributes are "equally important".
- DNaïve Bayes is (theoretically) based on probability theory.
18Information gain is ...
(select the correct answer!)- A... the "best" measure to select attributes in the process of building a decision tree.
- B... the information gained when classifying "new" examples with a decision tree.
- C... measured in bits.
- D... a measure for evaluating the performance of a decision tree.
19The entropy of having 4 equally likely events is:
(select the correct answer!)- A0
- B2
- C1
- D4
20The higher the Information gain ratio ...
- A... the better the attribute is in splitting the data into "pure" subsets.
- B... the worse the attribute is in splitting the data into "pure" subsets.
- C... the lower the Information gain of that attribute.
- D... the higher the Gini index of that attribute.
21How does C4.5 handle numeric attributes?
(select the correct answer!)- ABy ignoring them when building the decision tree.
- BBy discretizing them before building a decision tree.
- CBy searching for the best 2-way split (= splitting the data into 2 disjunct subsets).
- DBy using pruning.
22The higher the Gini index ...
- A... the better the attribute is in splitting the data into "pure" subsets.
- B... the worse the attribute is in splitting the data into "pure" subsets.
- C... the higher the Information gain of that attribute.
- D... the more data have missing values for that attribute.
23How does C4.5 handle missing values?
(select the correct answer!)- ABy removing them before building a decision tree.
- BBy using post-pruning.
- CBy incorporating them in the information gain ratio calculations in the form of weighted examples.
- DBy ignoring them when building the decision tree.
24What is pruning (of decision trees)?
(select the correct answer!)- AA sophisticated voting mechanism.
- BAn ensemble classification method involving more than one classification algorithm.
- CA method for evaluating the "quality" of decision trees.
- DA decision tree size reduction technique to avoid overfitting.
25Pre-pruning is ...
(select the incorrect answer!)- A... one of the mechanisms used by C4.5 to increase the robustness of the decision tree building process.
- B... (in practice) slower than post-pruning.
- C... often used to reduce the decision tree size and avoid overfitting the data.
- D... always faster than post-pruning.
26Post-pruning can ...
(select the correct answer!)- A... replace an entire sub-tree in a decision tree by a single leaf node.
- B... be used to handle missing values in the decision tree building process.
- C... always produce more accurate decision trees on the training set.
- D... be sometimes faster then pre-pruning.
27Select the only statement that is true!
- AA nominal attribute can appear more than once in a path from the root to a leaf node in a decision tree.
- BA one-level decision tree (just one attribute as the root node) produced by C4.5 is always identical to the model produced by OneR.
- CC4.5 always produces more accurate classifiers than Naïve Bayes.
- DC4.5 can use either pre-pruning, post-pruning, neither or both.
28Select the only statement that is false!
- AA numeric attribute can appear more than once in a path from the root to a leaf node in a decision tree.
- BC4.5 sometimes produces more accurate classifiers than Naïve Bayes.
- CA nominal attribute can appear more than once in a path from the root to a leaf node in a decision tree.
- DA decision tree "contains" as many decision rules as it has leaf nodes.
29J48 is ...
(select the correct answer!)- A... a decision tree building algorithm developed by the WEKA developers team, deemed superior to Quinlan's C4.5 algorithm.
- B... a JSON implementation of the well known 48th decision tree building algorithm.
- C... the J4.8 algorithm that Quinlan developed after his famous C4.5 algorithm.
- D... WEKA's Java implementation of Quinlan's algorithm C4.5 (revision 8).
30Decision rules "extracted" from a decision tree...
- A... may overlap.
- B... never overlap.
- C... sometimes don't overlap.
- D... always overlap.
31The PRISM algorithm produces ...
- A... a decision tree.
- Ban "ordered" set of rules that never overlap.
- Can "unordered" set of rules that may overlap.
- D... an instance-based classifier.
32Select the true statement:
- AThe PRISM algorithm is the only known algorithm to use the covering approach when learning classification rules.
- BThe PRISM algorithm supports both nominal and numeric attributes.
- CClassification rules are a form of IF...THEN rules for predicting a nominal class attribute.
- DThe covering approach is used by the C4.5 algorithm when building a decision tree.
33CN2 (like PRISM) is an algorithm that induces classification rules. The letters C and N stand for:
- AClassification using negation.
- BA Classifier Naive.
- CThe classifier for Numeric data.
- DThe initials of Clark and Niblett, the inventors of the algorithm.
34A covering approach "produces" decision rules...
- Athat may overlap
- Bthat never overlap
- Cthat can be always "merged" into a decision tree
- Dthat contain missing values
35An example e is said to "be covered" by a decision rule X->Y if
- AY(e) = true
- BY(e) = false
- CX(e) = true
- DX(e) = false
36Select the only false statement!
- AAll numeric attributes should be either discretized or removed before using the PRISM algorithm
- BThe PRISM algorithm can handle only nominal attributes
- CThe PRISM algorithm always produces an "unordered" rule set
- DDecision rules "extracted" from a decision tree sometimes overlap
37______ is always faster than ______.
- AJ48,PRISM
- Bpost-pruning, pre-pruning
- Cpre-pruning, post-pruning
- DOneR, ZeroR
38How does the WEKA implementation of the PRISM algorithm handle missing values?
- AIt uses an "example weighting scheme" similar to that of the J48 algorithm
- BIt simply ignores them
- CThe algorithm will not start, because it doesn't support missing values
- DIt uses class-dependent discretization (similarly to OneR)
39How does the WEKA implementation of the PRISM algorithm handle noisy data?
- AIt uses kind of pruning similar to the pre-pruning of the J48
- BIt post-prunes the rules using a sort of significance test
- CIt does not. The algorithm will overfit
- DIt uses class-dependent discretization (similarly to OneR)
40Unsupervised learning ...
- A... is used for decision tree construction in the C4.5 algorithm.
- B... finds groupings of instances given labeled data.
- C... always overlap.
- D... finds "natural" groupings of instances given unlabeled data.
41Select the only true statement!
- AIn unsupervised learning the data has to be always labelled.
- BThe PRISM algorithm performs unsupervised learning.
- CClustering is a form of unsupervised learning.
- DClustering is always a form of supervised learning.
42A model produced by a clustering algorithm ...
- A... always contains non-overlapping clusters of data.
- B... is always learnt in a top-down fashion.
- C... is always a decision tree.
- D... may contain overlapping clusters of data.
43In k-Means clustering ...
- A... "Means" is referred to average classification accuracy.
- B... we are guaranteed to get the "optimal" clusters, if the algorithm converges to a solution.
- C... the algorithm iteratively finds a solution, which may not be optimal.
- D... the algorithm always converges to an optimal solution.
44What is random restart in k-Means clustering?
(select the correct answer!)- AIt is a term that doesn't really apply to k-Means clustering, but rather to decision tree learning algorithms.
- BThe fact that the initial solution is chosen randomly.
- CMultiple re-starts of the algorithm with different random initializations.
- DRandom cluster centroid displacements during the iterations of the algorithm.
45Hierarchical agglomerative clustering ...
- A... is always performed in a "top-down" fashion.
- B... is a form of clustering where clusters are constructed in a "bottom-up" fashion.
- C... is very similar to k-Means clustering.
- D... is actually a decision rule learning algorithm.
46Select the only false statement!
- AIn agglomerative hierarchical clustering the number of clusters may be automatically detected.
- Bk-Means clustering is sensitive to the initial positioning of the cluster centroid points.
- CClustering is always a form of supervised learning.
- DC4.5 is a supervised learning algorithm.
47Why is it advisable to perform attribute normalization before performing any clustering algorithm?
(select the correct answer!)- ABecause with normalized attributes the clustering algorithms are more efficient.
- BBecause non-normalized attributes may cause clustering algorithms to crush.
- CBecause clustering can only be performed if all attributes are normalized.
- DBecause different value ranges of different attributes may bias the clustering process.
48What is "benchmarking on existing class labels"?
(select the correct answer!)- AIt is a form of performance test measuring the time efficiency of the clustering algorithms.
- BIt is a form of performance test measuring the memory efficiency of the clustering algorithms.
- CIt is when we perform clustering using just the class attribute and ignoring all the other attributes.
- DWhen we actually have labelled data, but perform clustering ignoring the class attribute and later compare the clusters with the class attribute classes.
49In data mining, if a model fits the training data to precisely, leading usually to poor results on new data, the phenomenon is called:
- ACross-validation
- BOverestimation
- COverfitting
- DNoise reduction
50Identify the one that is not a sample set used in data mining during the learning/evaluation process!
- AValidation set
- BTest set
- CPermutation set
- DTraining set
51When we have "enough" data, we typically evaluate our models using ...
- A... the random training/test split method.
- B... the bootstrap method.
- C... the cross-validation method.
- D... the "evaluate-on-training set" method.
52Balancing the data is needed when ...
- A... we don't have "enough" data.
- B... we have many strongly correlated attributes.
- C... the proportion of examples belonging to one class is much larger than the proportion of examples belonging to the other classes.
- D... we have a very large dataset.
53What is the typical number of folds for cross-validation?
- A20
- B5
- C10
- D2
54A validation set should be used when ...
- A... we have "insufficient" data.
- B... no test set is available.
- C... trying to identify the "best" set of parameter values for an algorithm.
- D... we are dealing with big data.
55What is a learning set?
- AA set of examples used to tune the parameter values for an algorithm.
- BA set of examples used to learn the model from.
- CA set of examples used to test the model on.
- DA "huge" set of examples with no class attribute.
56Classification accuracy is typically used ...
- A... as an evaluation measure for classification models.
- B... instead of the C4.5 algorithm for regression problems.
- C... to measure the correlation of results in a regression problem.
- D... as an evaluation measure for regression models.
57A test set is typically used ...
- A... just in case of big data.
- B... to evaluate a classifier's performance simulating new/unseen data.
- C... to identify the "best" set of parameter values for an algorithm.
- D... to learn a model from.
58Identify the one statement that is true for leave-one-out evaluation!
- AThe test set is guaranteed to be non-stratified.
- BStratification of training/test set can be switched on or off.
- CThe test set is always stratified.
- DThe training set is always stratified.
59If we deal with noisy data and use the k-NN algorithm, ...
- A... it is always good to reduce the value of the parameter k.
- B... it is always good to increase the value of the parameter k.
- C... it is best to use the setting k=1.
- D... this is always a bad choice, because k-NN does not work well on noisy data.
60In data mining, we speak about regression when ...
- A... the process of learning is unsupervised.
- B... we use cross-validation several times in succession for evaluating the model.
- C... we are learning association rules.
- D... we are trying to predict a numerical class.
61The method of nearest neighbors is one of the "lazy learning" classification methods. Why?
- ABecause researchers were lazy and the method was not fully developed.
- BBecause the method is very time-consuming.
- CBecause the method requires large amounts of memory.
- DBecause learning is postponed to the classification phase.
62In linear regression, the error that is most often minimized is:
- Athe relative error.
- Bthe absolute error.
- Cthe Manhattan error.
- Dthe square error.
63When using the method of nearest neighbors ...
- A... we always use the Euclidean distance to calculate the "proximity" between the examples.
- B... the distances between the values of the nominal attribute can be only 0 or 1.
- C... we always classify according to the class value of the nearest example.
- D... we cannot evaluate using the cross-validation method.
64Which of the following statements is incorrect?
- ALinear regression is the only possible regression used to model numerical variables.
- BThere are other types of regression besides linear regression.
- CRegression can also be used for classification.
- DRegression can be performed using an iterative procedure.
65What algorithm is obtained if the parameter k of the k-NN algorithm is set to n-1 (n = number of examples)?
- AZeroR.
- BOneR.
- CC4.5.
- DPRISM.
66What is (always) the range of values of any normalized attribute A?
- A[-1, +1].
- B[-average(A), +average(A)].
- C[0, 1].
- D[-variance(A), +variance(A)].
67Logistic regression ...
- A... is typically used to solve problems in logistics.
- B... is a sub-field of mathematical logic.
- C... is typically used to solve classification problems.
- D... is typically used to model numerical variables.
68What is linear regression?
- AConverting the exponential function to a linear one.
- BFinding a linear function of attributes that predicts the class by minimizing the error.
- CA linear function of the nth order.
- DFinding a model that best predicts the nominal class.
Información
- Descripción
- one Data mining Test sobre Sin título, creado por Nina Lazarevska el 29/01/2023.
- Autor
- Nina Lazarevska
- Idioma
- English
- Visitas
- 2
- Etiquetas
- Valoración
- Aún sin valoraciones Valora este test