AIM: Clustering the Data together

1
AIM Clustering the Data together

• Clustering not dependent upon visible features
• Visible features example
• fingerprints whorls/archs/loops
• These are dependent upon the genetic formation
• People of similar regions may have similarity in
  terms of these features
• Thus clustering based on such features is biased

2
Letting the Data speak

• Collect the data samples from the test domain
• Cluster the data based on some clustering
  algorithm

3
Clustering Algorithm used

• K-means clustering Algorithm
• Algorithm in short
• Start with a predefined number of clusters
• Initialize the clusters with a certain centroid
• Run the algorithm to associate each member point
  
• to a cluster
• Re-calculate the centroids

4
K-means clustering algorithm

1. Place K points into the space represented by the
   objects that are being clustered. These points
   represent initial group centroids.
2. Assign each object to the group that has the
   closest centroid.
3. When all objects have been assigned, recalculate
   the positions of the K centroids.
4. Repeat Steps 2 and 3 until the centroids no
   longer move. This produces a separation of the
   objects into groups from which the metric to be
   minimized can be calculated.

5
Algorithm Evaluation

• Non-intuitive, yet true
• Always Converges
• Reason finite number of ways of partitioning R
  into k groups
• Each time the configuration changes, we go to an
  improved distortion (sum of square error)
• Every iteration new configuration
• If run forever, number of configurations exhaust

6
Algorithm Evaluation

• Convergence does not guarantee Optimality
• To Assure a near-Optimal solution
• Careful selection of starting points
• Several runs of the Algorithm

7
Selecting the starting points

• Using the k-nearest neighbor concept
• k corresponds to the number of clusters
• Find the global mean of the entire dataset
• Find the 'k' - closest data points to the global
  mean
• These 'k'-closest data samples are the initial
  'k'- centroids.

8
The code so far.
9
References

• Statistical data mining Tutorial slides
• Andrew Moore
• (http//www-2.cs.cmu.edu/awm/tutorials/kmeans.htm
  l)
• A Tutorial on clustering algorithms
• (http//www.elet.polimi.it/upload/matteucc/Cluste
  ring/tutorial_html/kmeans.html)