You might like to include number of false "hits" as a negative metric?

There are a lot of metrics (see below), but what is suitable for you - depends of your task (customer requirements -:) :

1. FPR = FP/(FP + TN) - loss

2. FNR = FN/(FN + TP)

3. TPR = 1 - FNR (recall, sensitivity)

4. TNR = 1 - FPR (specifity) - garbage

For im-balanced data sets (i.e., amount of Pos

Did you try to compare the ROC curves of the algorithms?

• I compared between TOP-keywords + Biencube + TOPIC + TUBE + mine TAG

Please see the article published by me at Knowledge-Based Systems (Volume 22) Lakshminarayana Sadasivuni. Categorization of web pages – Performance enhancement to search engine. You can classify the results and then rank them

Regards

Lakshminarayana Sadasivuni

thanks for your answer Sir, But can you send me your article in my email : [email protected]