Performance Evaluation

Centre for Vision, Speech & Signal Processing

Classification Performance Evaluation

Since perfect classification cannot be achieved, it is necessary to peform an analysis of the classification error. The subject is split into two areas to deal with supervised and unsupervised classification, ie classification and clustering.

In the case of classification performance evaluation, classes are provided for:

The way in which a data set of samples is used to design or train the classifier and test its performance. These classes are derived from NumPerformanceC.
The pattern error function that determines the contribution of each sample of the test set to the estimate of the pobability of misclassification (PMC). These are derived from NumErrorC.

In the case of clustering, it is not possible to measure exactly how well the clustering has been performed since the correct answer is not known. However, there are a selection of clustering criteria which measure various features of the quality of the clusters that have been selected.

Subtopics:

AMMA
Pattern Recognition
Performance Evaluation
Implementation

Classes:

Default Classes	Description
NumErrorC	Handle class for classification error counting.
NumErrorBC	Implementation base class for pattern error functions.
NumPerformanceC	Handle class for classification performance measurement
NumPerformanceBC	Implementation base class for classification peformance evaluation.
NumClustCriterionC	Handle class for clustering criterion functions.
NumClustCriterionBC	Implementation base class for clustering criterion functions.

Author: Robert Crida, Created: 7/5/1998, Generated by DocEntry: March 20, 2001