• Menu
  About Myself Introduction Data Gathering Data Cleaning Exploring Data Clustering ARM and Networking Decision Trees Naive Bayes SVM Conclusions Infographic

Clustering

**Click here to get Record Data. Click here to get Text Data.

1) Clustering in Python: Click here to get Python code.

• Cluster visualizations of three different k-values

	k-value=2	k-value=3	k-value=4
For Record Data	(Click to see a larger view.)	(Click to see a larger view.)	(Click to see a larger view.)
For Text Data	(Click to see a larger view.)	(Click to see a larger view.)	(Click to see a larger view.)

For record data, the optimal k-value is 2. And for text data, the optimal k-value is 4.

• Dendrograms of three distance metrics

	Manhattan Distance	Euclidean Distance	Cosine Similarity
For Record Data	(Click to see a larger view.) The suggested k-value is 2.	(Click to see a larger view.) The suggested k-value is 2.	(Click to see a larger view.) The suggested k-value is 3.
For Text Data	(Click to see a larger view.) The suggested k-value is 4.	(Click to see a larger view.) The suggested k-value is 4.	(Click to see a larger view.) The suggested k-value is 5.

For both record and text data, Manhattan distance and Euclidean distance are more similar than Cosine similarity.

For record data, the optimal k-value is 2, and for text data, the optimal k-value is 4, which are the same as the previous results.

• 3D Visualizations

For Record Data	For Text Data
(Click to see a larger view.)	(Click to see a larger view.)

For record data, the x, y, z labels of the 3D visualization are the book categories. And for text data, the x, y, z labels of the 3D visualization are the most frequent words.

• Heatmaps

For Record Data	For Text Data
(Click to see a larger view.)	(Click to see a larger view.)

For record data, the category 0, which is computing book, has a much higher correlation. And for text data, the first document and the tenth document have the highest correlation.

2) Clustering in R: Click here to get R code.

• Three methods for finding good values of k

	Elbow	Silhouette	Gap Statistics
For Record Data	(Click to see a larger view.) The suggested k-value is 2.	(Click to see a larger view.) The suggested k-value is 2.	(Click to see a larger view.) The suggested k-value is 2.
For Text Data	(Click to see a larger view.) The suggested k-value is 1.	(Click to see a larger view.) The suggested k-value is 3.	(Click to see a larger view.) The suggested k-value is 1.

For record data, the optimal k-value is 2. For text data, the plots illustrate that 1 or 3 might be the optimal k-value. Thus, 3 clusters will be the best choice since k = 1 means a valueless clustering.

• Cluster visualizations of three different k-values

	k-value=2	k-value=3	k-value=4
For Record Data	(Click to see a larger view.)	(Click to see a larger view.)	(Click to see a larger view.)
For Text Data	(Click to see a larger view.)	(Click to see a larger view.)	(Click to see a larger view.)

It seems the optimal k-value is 2 for both record and text data.

• Hierarchical clustering plots of three distance metrics

	Manhattan Distance	Euclidean Distance	Canberra Distance
For Record Data	(Click to see a larger view.)	(Click to see a larger view.)	(Click to see a larger view.)
For Text Data	(Click to see a larger view.)	(Click to see a larger view.)	(Click to see a larger view.)

For record data, Manhattan distance and Euclidean distance are much more similar than Canberra distance. And for text data, Manhattan distance and Canberra distance are more similar.

For record data, the optimal k-value is 2, and for text data, the optimal k-value is 3, which are the same as the previous results.

• Visualizations of density clustering

For Record Data	For Text Data
(Click to see a larger view.)	(Click to see a larger view.)