K-Means Clustering

#3. Run KMeansNormTest.R with TestObservations and TestCenters. #a) What is the single most obvious difference between the distributions of the first and second dimensions?

The sizes of the dimensions are different, i.e. the data is not in equal terms.

#b) Does clustering in Test 1 occur along one or two dimensions? Which dimensions? Why?

In Test 1, clustering occurs in one dimension, the second dimension since it has a larger scale/impact than first dimension.

#c) Does clustering in Test 2 occur along one or two dimensions? Which dimensions? Why?

In Test 2, clustering occurs in one dimension, the second dimension. Normalizing the first dimension makes its scale even more smaller compare to the second dimension.

#d) Does clustering in Test 3 occur along one or two dimensions? Which dimensions? Why?

In Test 3, clustering occurs in two dimensions. Even though the second dimension has a larger scale, de-normalizing reduced its impact and cluster in two dimensions.

#e) Does clustering in Test 4 occur along one or two dimensions? Which dimensions? Why?

In Test 4, clustering occurs in two dimensions. Normalizing both dimensions balanced the impact of each dimension, i.e. it put the data in equal terms.

#4. Why is normalization important in K-means clustering?

It carries both cluster at an equivalent scale so that the impact of each dimension can be comparable in K-means clustering.

#5.How do you encode categorical data in a K-means clustering?

We need to transfer categorical values into numerical values, i.e. apply Binarization to be able to perform K-means clustering.

#One way to do that would be define impact values of each category (for instance the number of repetition of each category) . #6. Why is clustering un-supervised learning as opposed to supervised learning?

Unlike supervised learning which requires an expert label (or a specific function to guess dependent y via independent x), un-supervise learning users do not know what the outcome is.

ozemreozdemir / k_means_clustering Goto Github PK