University of Konstanz
Graduiertenkolleg / PhD Program
Computer and Information Science

Colloquium of the Department and the PhD Program


Building better data mining models with ROC analysis


Prof. Dr. Peter Flach, University of Bristol
Bristol, UK

date & place

Wednesday, 25.01.2006, 16:15 h
Room C252


Classification is a classical data mining problem where the value of a discrete (often binary) dependent "class" variable must be predicted from the values of the independent variables. Many data mining models, such as naive Bayes and decision trees, model a probability distribution over the dependent variable from training data, and then predict whichever class is estimated to have the highest probability. Receiver Operating Characteristic analysis, a decision making model originating from signal detection theory, can be used to obtain more sophisticated decision rules which result in better data mining models (or rather, which make better use of the models). In this talk I will overview some of the more innovative applications of ROC analysis in data mining.