CAP4770 Exam 1

Classify bronze, silver, and gold medals as awarded at the Olympics:

• binary

• discrete

• ordinal

• continuous

Ordinal - categories with meaningful order/ranking

Predicting sales amounts of a new product based on advertising expenditure is an example of regression. T/F?

TRUE

“Outlier” can be desired or interesing. T/F?

TRUE

Binary attributes are a special case of categorical attributes. T/F?

FALSE (special case of discrete)

Combining two or more attributes (or objects) into a single attribute (or object) is an example of …

• sampling

• aggregation

• feature extraction

• feature selection

Aggregation

In sampling without replacement, the same object can be picked up more than once. T/F?

FALSE

In order to do simple random sampling, we split the data into several partitions; and draw random samples from each partition. T/F?

FALSE

Arrangement is a broader term that encompasses both similarity and dissimilarity. It’s like having a single word to describe how two things are connected - whether they’re alike or different. T/F?

FALSE

Euclidean Density: Center-Based is one of the simplest approaches to divide the region into a number of rectangular cells of equal volume and define density as a number of points teh cell contains. T/F?

FALSE

Jaccard similarity = number of 11 matches / number of non-zero attributes

x = 0101010001

y = 0100011001

What is the Jaccard similarity?

0.6

For a coin with probability p of heads and probability q =1 - p of tails.

For p = 1, what is the entropy (H)?

• 0

• 1

• 1.32

• 2

0

While correlation is a valuable measure for assessing linear relationships, it’s important to remember that it only captures linear connections. In cases where the relationship between variables is non-linear, the correlation coefficient may not accurately represent the underlying patter. T/F?

TRUE

Cosine similarity measures the cosine of the angle between two vectors and is especially useful for text-based data. T/F?

TRUE

When you are the fourth tallest person in a group of 20, it means that you are the 80th percentile. T/F?

TRUE

Assume we have three points A, B, and C as shown in the figure. No matter which distance metric we employ, we can assert that the distance from A to C is larger than the distance from A to B. T/F?

FALSE

Minkowski distance with r=1 is Manhattan distance. T/F?

TRUE

Entropy quantifies the amount of information needed to describe the state of a system or the unpredictability of an outcome. T/F?

TRUE

EDA is a short form of Exploratory Data Analysis. T/F?

TRUE

Density measures the degree to which data objects are close to each other in a specified area. Concept of density is typically used for clustering and anomaly detection. T/F?

TRUE

The curse of dimensionality means that adding more variables always improves the performance of data models. T/F?

FALSE

PCA is suitable for both numerical and categorical variables. T/F?

FALSE

PCA focuses on maximizing the variance in data. T/F?

TRUE

LDA is an unsupervised dimensionality reduction technique. T/F?

FALSE

t-SNE is primarily used for feature reduction before classification. T/F?

FALSE

In PCA, the principal components are orthogonal to each other. T/F?

TRUE

LDA aims to maximize the variance between classes and minimize the variance within classes. T/F?

TRUE

The maximum number of principal components in a dataset with n samples and p features is min(n-1, p). T/F?

TRUE

PCA always leads to perfect class separability. T/F?

FALSE

t-SNE captures non-linear relationships in data better than PCA and LDA. T/F?

TRUE

Which of the following best describes the curse of dimensionality?

• adding more variables leads to sparse data space and poor model performance

• adding more variables improves model performance

• adding more data points always solves data sparsity

• none of the above

Adding more variables leads to sparse data space and poor model performance

Which dimensionality reduction technique is supervised?

• t-SNE

• LDA

• PCA

• All are supervised

LDA

Which technique focuses on maximizing the variance in the data?

• Logistic Regression

• t-SNE

• LDA

• PCA

PCA

Which technique is most suitable for visualizing high-dimensional non-linear data?

• Random forest

• LDA

• PCA

• t-SNE

t-SNE

What is the main goal of LDA?

• Maximize variance in data

• Minimize error in data points

• Maximize class separability

• Reduce number of features randomly

Maximize class separability

Which of the following steps is common to both PCA and LDA?

• Reducing dimensions based on neighborhood probabilities

• Standardizing the data

• Using class labels reduction

• Calculating between-class scatter

Standardizing the data

t-SNE is particularly useful for:

• Visualizing high-dimensional data

• Linear transformations of data

• Optimizing regression models

• Feature selection for classification

Visualizing high-dimensional data

In PCA, the eigenvector with the highest eigenvalue is called:

• The main axis

• The principal component

• The variance vector

• The discriminant function

The principal component

Which of the following is a limitation of PCA?

• PCA doesn’t have any limitations

• It is supervised

• It ignores class labels

• It requires non-linear separability

It ignores class labels

Which technique focuses on maximizing the separation between classes while minimizing variance within each class?

• LDA

• PCA

• t-SNE

• Factor analysis

LDA

What are the categorical attributes?

Nominal: no coherent order

Ordinal: meaningful order or rank

What are the numerical attributes?

Interval: meaningful order and consistent intervals

Ratio: meaningful order and a TRUE ZERO

What is a discrete attribute?

Has a finite or countably infinite set of values

BINARY is a special case of DISCRETE

What is a continuous attribute?

has real numbers as attribute values

(typically represented as floating point values)

How do we identify outliers?

Z-score; this measures how many standard deviations a data point is away from the mean of a dataset

Can outliers have a positive impact too?

Yes; they can provide insight on unusual trends, opportunities, or exceptional cases in data.

What is clustering and why is it used?

Clustering finds groups of objects with similar relation to one another within each group compared to other groups.

Used for understanding groups of similarities in datasets and reducing the size of large datasets.

What is Association Rule Discovery?

Finding similarities or relations between variables in databases.

Consider a dataset where the “Income” attribute is heavily skewed. What transformation technique would you apply to normalize this data, and why?

Logarithmic transformation; this will reduce skewness by normalizing the data distribution