作者：Graeme Smith 16 年以前
308
Classifier Evaluation Presentation
The text outlines a structured approach to evaluating classifiers, beginning with an introduction that sets objectives and methods for comparison. It delves into the intricacies of black box classifiers, detailing the stages and features necessary for effective scoring and selection.
開啟
Classifier EvaluationPresentation 8. What Comes Next? How to predict performance Advert for my Australia paper Possibly use error bounds Why do this? How to use "Don't Declare" threshold 7. Summary What have I told them about 6. Real World Examples Real world ROC curves Using a target level of Pfa Impact of a very cautious threshold Impact of forced decision Pre-processing and classifier used Introduce dataset 5. ROC Curve Analysis Interesting points in ROC space Generating the curve Type (b) ROC curve Type (a) ROC curve 4. The Confusion Matrix Test unknown target matrix Test variant target matrix Test target matrix 3. Evaluation Metrics Probability of generalization Probability of false alarm Probability of declaration Forced declaration Reliability Probability of correct classification 2. Black Box Classifier Outputs "Don't Declare" Unknown Declaration Stages Selection "Don't declare"
Declare
Threshold Scoring Features Inputs Data to classify Reference 1. Introduction Objectives Metrics Comparison Method Radar ATR
