I am using pyod for anomaly detection and for different algorithms,I am getting different no of outliers at the same contamination level for example abod is giving 5 outliers and IForest is giving 9 outliers at a contamination level of 0.05.
In the documentation I read like no of outliers are equal to total no of samples * contamination,then the number of outliers from all the algorithms at the same contamination level should be equal right,please suggest me on this.
Asked
Active
Viewed 67 times
0
learner
- 1
-
Is this when you evaluate on your training set? – Jon Nordby Feb 08 '22 at 16:12
-
How many samples in your training set? – Jon Nordby Feb 08 '22 at 16:12
-
@JonNordby,Thanks for the response,my training set consists of 180 samples and I did fit the algorithms on the training data and got the anomalies by using classifier.labels_. However I observed that if scores are same for more than mentioned percentage of outliers,then this is happening ,in my case after the scores of five anomalies, there are 7 data points with same anomaly score, since it is not possible to select the 4 data points out of 7 as they have same anomaly score, only the top five are getting returned. This is only my assumption, please correct me if I am wrong. – learner Feb 09 '22 at 17:39