3. PERFORMANCE ANALYSIS TOOLS › 3.3 Data Clustering Analysis › 3.3.4 Analytic Technique Tutorial › 3.3.4.3 Algorithm Implementation

3.3.4.3 Algorithm Implementation

The Data Clustering program is comprised of these major
steps:

 1 - Select data elements (features) from the CA MICS file
     based on user input.

 2 - Execute Neugents technology training phase, which
     examines the input data and builds a set of clusters
     using its own proprietary techniques.

 3 - Neugents technology then executes the "consulting"
     phase, where all observations (patterns) are consulted
     against the trained model and each observation or
     pattern is assigned to its final cluster.  Outliers are
     marginally assigned during this process.

 4 - Clustered data, along with model performance statistics
     are returned to the user.


The first step in the program is the selection of the user-
specified elements from a CA MICS database file.  The file
and database elements (features) are specified with
parameters on the Data Clustering screen.  The control
parameters are discussed in the sections that follow.

The second step is to perform the training phase, in a manner
not unlike that done in the Workload Forecasting.  A model of
clusters is produced from the input data using an iterative
process involving a combination of hierarchical and "k-means"
methods.  The actual process involved is proprietary, but
permits accurate and rapid cluster creation.  A Relative
Position value is calculated for each observation or pattern
in the manner previously described for cluster evaluation
purposes, and the model is then ready for the final phase.

During the next phase, the input data is reprocessed and
"consulted" against the trained model.  Each observation is
now assigned to a permanent cluster and any outliers are
marginally assigned, meaning they are put with a cluster that
best describes the vector data contents.

Lastly, the final clustered data, along with model
performance statistics is made available to the Data
Clustering application.