Question

1. What is clustering? Explain how K-Means Clustering Algorithm works?

2. What are the Advantages and disadvantages of Clustering ALgorithms discussed in our class (K-Means,Hierchal)?

3. Which Clustering Algorithm is better K-Means or hierarchical Clustering? Explain with a proper example which is better algorithm?

Answer 1

What is clustering:

A cluster refers to a collection of data points aggregated together because of certain similarities.
Data clustering approaches can group similar data into clusters. The grouped Data will usually reveal important meanings.
Data that are close to each other tend to share some external relationship. This relationship can be established to group the data into clusters.

How K-Means Clustering Algorithm works:

• In k∗-means, we also use the random initialization method to choose k∗ starting centers, and first assign all points into k∗clusters.
• Then we get feature values of mean for each cluster,as shown in lines 1-3. Next, k∗-means performs hierarchical clustering along with k-means adjusting iteration in lines 4-22 and nearest clusters associated with top-n distances merging in line 23-25.
• Line 6 describe the proposed cluster pruning strategy. We use collections CS to store the neighbor clusters of specific cluster which after prune remote clusters by Lemma 2.
• Then, we only need to verify the adjustable clusters in search space of CS for each point in Ci.
• Once percentage of moved points is lower than given θ in first round k-means optimized update principle will be started and radius will be updated during this process( lines 7- 14).
• For algorithm’s efficiency, we maintain a value r in each cluster to update its radius( lines 9-13). At the end of each iteration, each cluster mean m and it’s radius is replaced by , directly.Lines 23-25 show top-n nearest clusters merging which reduce number of clusters from
• Therefore, parameter n is not fixed, but ranges from given n to 1. For each round refining of k∗-means, we use a decrease strategy to determine value of n, and a top-1 may be performed at final round to make number of clusters reach at k.

----------------------------------------------------------------------------------------------------------------------