Browse Prior Art Database

Illustrating Confidence Measures Graphically for a Segmentation Model in Data Mining

IP.com Disclosure Number: IPCOM000020409D
Original Publication Date: 2003-Nov-20
Included in the Prior Art Database: 2003-Nov-20
Document File: 1 page(s) / 5K

Publishing Venue

IBM

Abstract

Illustrating confidence measures graphically for a segmentation model in data mining

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 100% of the total text.

Page 1 of 1

Illustrating Confidence Measures Graphically for a Segmentation Model in Data Mining

During the model build process a set of histogram bins are maintained by the system. These bins are displayed graphically at the end of each training pass allowing the analyst to understand the distribution of confidence measures inside each cluster or segment in the model. Traditional data mining programs rely on only displaying a score or error value for the closeness of a cluster in the model. This method in addition to the score displays an easy to view distribution of scores, it is easy to understand how tightly bound the scores are inside each cluster, where as a single number does not obviously provide the same visual impact. Equally, splits in the distribution can have important implication in interpreting the results of the clustering. The total range of scores is 0 to 1 this is divided into ten even cells and the count of records falling into each range is maintained. The output is displayed as a histogram. This uses a graphical method rather than a single number.

1