Final Exam: Predictive Analytics

Question 1

What is Clustering?

Answer 1

The process of grouping a set of data points into Clusters.

Question 2

What is Intra-Cluster Distance?

Answer 2

Data within the same cluster will be similar

Question 3

What is inter-cluster distance?

Answer 3

Data from different clusters will be different.

Question 4

What are the inputs and outputs of clustering?

Answer 4

Input: Unlabeled data.
Output: Assign each data point to a cluster.

Question 5

Is Clustering Supervised or Unsupervised Learning?

Answer 5

Unsupervised Learning

Question 6

What are the motivations of clustering?

Answer 6

1. Clustering breaks a large heterogeneous population into homogeneous subgroups
2. Common step in the exploratory data analysis (EDA) to gain insights
3. Useful to reduce data dimension when analyzing high-dimensional data

Question 7

What are some applications of clustering?

Answer 7

Targeted advertising, recommendation, and fraud detection.

Question 8

What is k-means clustering?

Answer 8

A clustering algorithm that partitions a data set into K distinct, non-overlapping clusters.

Question 9

What is the process for K-Means algorithm?

Answer 9

The process that involves iteratively updating the cluster centroids and group assignments.
1. Randomly assign each point to one of the K clusters
2. Iterate until (1) no change in assignments or (2) reaching the pre-set max number of iterations

Question 10

What is the Euclidean Distance?

Answer 10

The distance between the k-th point and l-th point is xk - xl squared plus other iteration underneath the square root of the total.

Question 11

What are some practical details to K-means clustering?

Answer 11

1. K is a hyperparameter, which is pre-set before running the algorithm.
2. Results may vary based on random initial centroid selection. Run the algorithm multiple times and select the most reasonable result
3. A good clustering produces high-quality clusters where (1) Intra-cluster distance is small (2) Inter-cluster distance is large (3) Information gained is meaningful and can be applied

Question 12

What are some limitations to k-means clustering?

Answer 12

1. Different sizes
2. Different density
3. Non-spherical shapes

Question 13

What is hierarchical clustering?

Answer 13

The process that results in an attractive tree-based representation of the data points, called dendrogram.

Question 14

What are the characteristics of a dendrogram?

Answer 14

1. A dendrogram is built (interpreted) starting from the leaves in the bottom.
2. Each leaf contains one data point
3. Moving up, some leaves fuse into branches
4. Fused leaves are “similar” to each other
5. The earlier (lower in the tree) fusions occur, the more similar the leaves are

Question 15

How to choose the number of clusters in a dendrogram?

Answer 15

Make a horizontal cut across the dendrogram.

Question 16

What are some practical details of hierarchical clustering?

Answer 16

1. Similar to K-means, we also need to configure the intra-cluster distance using euclidean/correlation and the inter-cluster distance (linkage) using complete, single, and average
2. Part of EDA, Understand the implications of the cut, and choose the one that serves the best for the core problem.

Question 17

What key attributes are we investigating in regards to clusters in RapidMiner?

Answer 17

1. Average with cluster distance: inter-cluster
2. Davies Bouldin Index (DBI): A combination of intra- and inter- cluster distances (The smaller, the better)
3. The purity of each cluster according to a certain variable (The purer, the better)

Question 18

What is a support vector machine?

Answer 18

A supervised learning algorithm that was developed for classification problems and is often one of the best “out of the box” classifiers.

Question 19

What is the intuition behind a support vector machine?

Answer 19

To look for a hyperplane separating two classes.

Question 20

What is a hyperplane?

Answer 20

A decision boundary used for a classification and regression.
f(x1, x2) = B0 + B1x1 + B2x2

Question 21

What is a prediction for a hyperplane observation?

Answer 21

Predictions are made by checking which side of the hyperplane a point falls on.
If f(xi1, xi2) is greater than or equal to 0, then y = 1.
If f(xi1, xi2) is less than 0, then y = -1.

Question 22

What is machine learning?

Answer 22

The science (art) of programming computers so they can learn from data, and generalize/predict unknown information.

Question 23

What is unsupervised learning?

Answer 23

Data does that does NOT contain labels.
Goal: Understand data pattern.

Question 24

What is supervised learning?

Answer 24

Data that contains labels.
Goal: Predict label.

Question 25

Examples of supervised learning?

Answer 25

1. Detect tumors in brain scans: yes or no? 2. Recognize a handwriting number: 0, 1, ....., 9? 3. Forecast your company's revenue next year, based on many performance metrics

Question 26

Examples of unsupervised learning?

Answer 26

1. Segment customers based on their purchases so that you can design a different marketing strategy for each segment. 2. Represent a high-dimensional dataset in a clear and insightful diagram 3. Understand the associations between products, such as how likely a customer is to buy milk after buying bread

Question 27

What is regression?

Answer 27

A supervised learning model utilized when the label variable is numerical.

Question 28

What is classification?

Answer 28

A supervised learning model when the label variable is categorial.

Question 29

What does the Mean Squared Error tell us?

Answer 29

The average of the squared differences between predicted and actual values, providing a measure of how well a model's predictions align with the real data, with lower values indicating better model fit.

Question 30

What is training data?

Answer 30

Data utilized to train the prediction model?

Question 31

What is test data?

Answer 31

Data utilized to evaluate the accuracy, precision, and recall prediction model using new data that is "fresh and unseen"

Question 32

What kind of model is the Right Fit for General Data?

Answer 32

A model where Training is Low and Test is Low.

Question 33

What kind of model is an overfit for the general data?

Answer 33

A model where Training is Low and Test is High.

Question 34

What kind of model is an underfit for the general data?

Answer 34

A model where Training is High and Test is Low.

Question 35

What is a decision tree?

Answer 35

A decision is a popular and versatile prediction model that can perform both classification and regression tasks.

Question 36

Examples of decision nodes?

Answer 36

A root node and internal node.

Question 37

What are branches?

Answer 37

They are lines connecting nodes that specify the possible outcomes.

Question 38

What are leaf nodes?

Answer 38

The model's predictions.

Question 39

What is a confusion matrix?

Answer 39

A table that summarizes the performance of a classification model by comparing its predictions against the actual values.

Question 40

What is accuracy?

Answer 40

The percentage of predictions that match the actual labels. Equation: True Positives + True Negative/The Entire Dataset

Question 41

What is precision?

Answer 41

The percentage of true positives identified from a model's predicted positives. Equation: TP/TP + FP

Question 42

What is recall?

Answer 42

The percentage of true positives identified from a model's actual positives. Equation: TP/TP + FN

Question 43

How would you describe precision to someone with no technical background?

Answer 43

When I predict _________, 98% of my predictions are truly __________. When I predict _________, 94% of my predictions are truly __________.

Question 44

How would you describe recall to someone with no technical background?

Answer 44

Among all the acceptable cars, my prediction model can identify 92% of them.

Question 45

What does R Squared tell us in this dataset?

Answer 45

How well a model fits a dataset. In other words, it determines whether a model has strong or weak predictive powers