Browse Prior Art Database

Evaluation of Predictive Models Produced by Datamining

IP.com Disclosure Number: IPCOM000014689D
Original Publication Date: 2000-Jun-01
Included in the Prior Art Database: 2003-Jun-20
Document File: 2 page(s) / 231K

Publishing Venue

IBM

Abstract

Disclosed is a method for quickly visualizing and evaluating the results of a predictive model against actual data. This method graphs the predicted results and range of actual results. The user can then quickly evaluate the accuracy of the model relative to the groups of interest. How to read graph: The graph is shown in "decile" format. Customers are ranked based on their predicted results. This ranking is then subdivided into ten groups (deciles). 1

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 100% of the total text.

Page 1 of 2

Evaluation of Predictive Models Produced by Datamining

Disclosed is a method for quickly visualizing and evaluating the results of a predictive model against actual data. This method graphs the predicted results and range of actual results. The user can then quickly evaluate the accuracy of the model relative to the groups of interest.

How to read graph:

The graph is shown in "decile" format. Customers are ranked based on their predicted results. This ranking is then subdivided into ten groups (deciles).

1

[This page contains 248 pictures or other non-text objects]

Page 2 of 2

Each decile contains an "I-bar". Values for the I-bar are computed from actual data. The center of the I-bar is placed at the average value for that decile. The length of the I-bar is based on the standard deviation. For this graph, at +/- 1 standard deviation.

A continuous, solid step-function line spans all deciles. This solid line represents results produced by the predictive model and represent the average value predicted for that decile.

The quality of the model is quickly and easily determined by evaluating how well the predicted model fits within the range of actual values for the deciles of interest.

2