Week 4: Clustering 1

Question 1

What is the formula for Euclidean distance?

Answer 1

Square root of the sum of (Ai - Bi)^2

Question 2

What are the 2 main types of clustering?

Answer 2

1) Hierarchical - repeatedly combine the 2 closest clusters (or points, during the first run) to become 1 new cluster. Usually ends up with a few big clusters of points
2) Partitioning method - maintain a set number of clusters, and fit points into the cluster closest to them

Question 3

What exactly is the main purpose of setting K number of unobserved clusters?

Answer 3

It is to remove any identifying labels of the points, such that there will be no bias + you might find answers you previously could not see

Question 4

When do you stop the K-means algorithm?

Answer 4

When the change in SSE is small

Question 5

What are some challenges of the K-means algo?

Answer 5

• Minimising SSE is computationally very difficult.
• K-means algorithm is not guaranteed to converge to the optimal solution, and different starting seeds might produce different solutions

Because of these problems, analysts might want to estimate the K-means solution multiple times with different starting seeds and select the solution with the smallest value of SSE.

Question 6

Since distance measure can only be used for numerical variables, how do we use it for categorical variables?

Answer 6

• Binary or Two-value variable (0-1 variable)
• (k-1) variables dummies where k = number of categories in the variable