28 กรกฎาคม 2553 4:24
In clustering, how does Analysis Services determine which value belong to each class?
For example, here:
If a set of input of features is classified 51% to be "0" and 49% to be "1", does it mean that Analysis Services consider the input as "0"?
28 กรกฎาคม 2553 5:50ผู้ดูแล
In all predictive algorithms, the classification matrix will consider the most likely value to be the predicted value (therefore, the value yielding the highest probability).
If you want to estimate the model performance for different probability thresholds, you should use the Lift Chart (well, actually cumulative gains chart) included in the product. You can see in that chart, in your example, how many correct "1"'s are being found by Analysis Services if your probability threshold were 30% (instead of 50%)
bogdan crivat / http://www.bogdancrivat.net/dm
2 พฤษภาคม 2555 15:06how do you generate lift chart with different threshold?
2 พฤษภาคม 2555 15:53is there any validation process for estimate the performance of clustering?
Acoolme :Anybody Can Be Your Marketer