slide 1:

3/22/2012 1 K-means Algorithm g Cluster Analysis in Data Mining Presented by Zijun Zhang Algorithm Description  What is Cluster Analysis Cluster analysis groups data objects based only on information found in data that describes the objects and their relationships. Goal of Cluster Analysis The objects within a group be similar to one another and jgp different from the objects in other groups

slide 2:

3/22/2012 2 Algorithm Description  Types of Clustering Partitioning and Hierarchical Clustering  Hierarchical Clustering - A set of nested clusters organized as a hierarchical tree  Partitioning Clustering gg - A division data objects into non-overlapping subsets clusters such that each data object is in exactly one subset Algorithm Description p4 p1 p3 p2 A Partitional Clustering Hierarchical Clustering

slide 3:

3/22/2012 3 Algorithm Description  What is K-means 1. Partitional clustering approach 2. Each cluster is associated with a centroid center point 3. Each point is assigned to the cluster with the closest centroid 4 Number of clusters K must be specified 4. Number of clusters K must be specified Algorithm Statement  Basic Algorithm of K-means

slide 4:

3/22/2012 4 Algorithm Statement  Details of K-means 1 Initial centroids are often chosen randomly 1. Initial centroids are often chosen randomly. - Clusters produced vary from one run to another 2. The centroid is typically the mean of the points in the cluster. 3.‘Closeness’ is measured by Euclidean distance cosine similarity correlation etc. 4. K-means will converge for common similarity measures mentioned above. 5. Most of the convergence happens in the first few iterations. 5. Most of the convergence happens in the first few iterations. - Often the stopping condition is changed to ‘Until relatively few points change clusters’ Algorithm Statement  Euclidean Distance A simple example: Find the distance between two points the original and the point 34

slide 5:

3/22/2012 5 Algorithm Statement  Update Centroid We use the following equation to calculate the n dimensional We use the following equation to calculate the n dimensional centroid point amid k n-dimensional points Example: Find the centroid of 3 2D points 24 52 and 89 and 89 Example of K-means  Select three initial centroids 1 1.5 2 2.5 3 y Iteration 1 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 x

slide 6:

3/22/2012 6 Example of K-means  Assigning the points to nearest K clusters and re-compute the centroids 1 1.5 2 2.5 3 y Iteration 3 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 x Example of K-means  K-means terminates since the centroids converge to certain points and do not change. 1 1.5 2 2.5 3 y Iteration 6 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 x

slide 7:

3/22/2012 7 Example of K-means 2 2.5 3 Iteration 1 2 2.5 3 Iteration 2 2 2.5 3 Iteration 3 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 x y -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 x y -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 x y 3 Iteration 4 3 Iteration 5 3 Iteration 6 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 2 2.5 x y -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 2 2.5 x y -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0 0.5 1 1.5 2 2.5 x y Example of K-means  Demo of K-means

slide 8:

3/22/2012 8 Evaluating K-means Clusters  Most common measure is Sum of Squared Error SSE  For each point the error is the distance to the nearest cluster  To get SSE we square these errors and sum them  To get SSE we square these errors and sum them.  x is a data point in cluster Ci and mi is the representative point for cluster Ci  can show that mi corresponds to the center mean of the cluster  Given two clusters we can choose the one with the smallest error    K iC x i i x m dist SSE 1 2  Given two clusters we can choose the one with the smallest error  One easy way to reduce SSE is to increase K the number of clusters  A good clustering with smaller K can have a lower SSE than a poor clustering with higher K Problem about K  How to choose K 1. Use another clustering method like EM. 2. Run algorithm on data with several different values of K. 3. Use the prior knowledge about the characteristics of the problem.

slide 9:

3/22/2012 9 Problem about initialize centers  How to initialize centers - Random Points in Feature Space - Random Points From Data Set - Look For Dense Regions of Space - Space them uniformly around the feature space Cluster Quality

slide 10:

3/22/2012 10 Cluster Quality Limitation of K-means  K-means has problems when clusters are of differing g  Sizes  Densities  Non-globular shapes K hbl h hd i  K-means has problems when the data contains outliers.

slide 11:

3/22/2012 11 Limitation of K-means Original Points K-means 3 Clusters Application of K-means  Image Segmentation The k-means clustering algorithm is commonly used in computer vision as a form of image segmentation. The results of the segmentation are used to aid border detection and object recognition.

slide 12:

3/22/2012 12 K-means in Wind Energy  Clustering can be applied to detect b lit i i d d t b l abnormality in wind data abnormal vibration  Monitor Wind Turbine Conditions  Beneficial to preventative maintenance  K means can be more powerful and  K-means can be more powerful and applicable after appropriate modifications K-means in Wind Energy Modified K-means

slide 13:

3/22/2012 13 K-means in Wind Energy  Clustering cost function 2 1 1 ji k ji iC dk n        x xc x c 1 k i i nm    2 1 k    1 1 1 ji ji k iC i i dk m          x xc x c K-means in Wind Energy  Determination of k value 002 0.03 0.04 0.05 0.06 0.07 0.08 0.09 Cost of clustering 0 0.01 0.02 23 456789 10 11 12 13 Number of clusters

slide 14:

3/22/2012 14 K-means in Wind Energy  Summary of clustering result No. of Cluster c 1 Drive train acc. c 2 Wind speed Number of points Percentage 1 71.9612 9.97514 313 8.75524 2 65.8387 9.42031 295 8.25175 3 233.9184 9.57990 96 2.68531 4 17.4187 7.13375 240 6.71329 5 3.3706 8.99211 437 12.22378 6 0.3741 0.40378 217 6.06993 7 18.1361 8.09900 410 11.46853 8 0.7684 10.56663 419 11.72028 9 62.0493 8.81445 283 7.91608 10 81.7522 10.67867 181 5.06294 11 83.8067 8.10663 101 2.82517 12 0.9283 9.78571 583 16.30769 K-means in Wind Energy  Visualization of monitoring result

slide 15:

3/22/2012 15 K-means in Wind Energy  Visualization of vibration under normal condition 14 4 6 8 10 12 14 Wind speed m/s 0 2 0 20 40 60 80 100 120 140 Drive train acceleration Reference 1. Introduction to Data Mining P.N. Tan M. Steinbach V. Kumar Addison Wesley 2. An efficient k-means clustering algorithm: Analysis and implementation T. Kanungo D. M. Mount N. Netanyahu C. Piatko R. Silverman and A. Y. Wu IEEE Trans. PatternAnalysis and Machine Intelligence 24 2002 881-892 3. http://www.cs.cmu.edu/cga/ai-course/kmeans.pdf 4. http://www.cse.msstate.edu/url/teaching/CSE6633Fall08/lec1620k-means.pdf

slide 16:

3/22/2012 16 Appendix One Original Points K-means 2 Clusters Appendix Two Original Points K-means Clusters One solution is to use many clusters. Find parts of clusters but need to put together.