K-means clustering

Characteristics of the Input Data are Important

Dimensionality

• Noise and Outliers

• Type of Distribution

• Type of Data / Attributes – dictates type of similarity

Data: Pre-processing

– Normalize Data

– Eliminate Outliers

Data: Post-processing

– Eliminate small clusters that may represent outliers

– Split “loose” clusters; i.e., clusters with relatively high SSE

– Merge clusters that are “close” and that have relatively low SSE

– Can use these steps during the clustering process

K – Means Clustering

Given a value of k, the k-means algorithm randomly assigns each observation to one of the k clusters.

• After all observations have been assigned to a cluster, the resulting cluster centroids are calculated.

• Using the updated cluster centroids, all observations are reassigned to the cluster with the closest centroid

How to choose “k”?

Choose “k” based on how results will be used

– Example: “How many market segments do we want?”

• Also, experiment with slightly different k’s

If the no of clusters, k, is not clearly established by the context of the business problem, the k-means algorithm can be repeated for several values of k to identify promising values.

Suitability of k-Means Clustering

Suitable when you know how many clusters you want and you have a larger data set (e.g., more than 500 observations)

• This method is appropriate for larger tables upto millions of rows and allows only numerical data.

• Partitions the observations, which is appropriate if trying to summarize the data with k “average” observations that describe the data with the minimum amount of error.

Clustering should result in groups….

made of observations that are more similar too each other than they are to observations in other clusters.

Cluster cohesion

relates to the distance between observations within the same cluster.

Cluster separation

relates to the distance between observations in different clusters.

Cluster interpretability

relating to how much insight clusters provide.

Cluster stability

referring to how robust is the set of clusters with respect to slight changes in the data

Clustering is an…

UNSUPERVISED TECHNIQUE