Evaluate Replicator Neural Network for Anomalydetection

Question

Marvin Klatte 2021-1-26

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/727108-evaluate-replicator-neural-network-for-anomalydetection

编辑： Jayant Gangwar 2022-10-4

Hi,

i am trying to evaluate a Replicator Neural Network for Anomalydetection. For training i am using "normal" Data, without any Anomalies. For Evaluation i only have "normal" Data as well. Thats why i cant use (ROC,AUC....). Do you have any Idea how to evaluate Anomaly-Detection without Anomalies in Evaluation/Testing Data?

I appreciate your help. Thanks

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Jayant Gangwar 2022-10-4

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/727108-evaluate-replicator-neural-network-for-anomalydetection#answer_1066180

编辑：Jayant Gangwar 2022-10-4

Hi Marvin,

You can use parameters like precision, recall, F1 score to evaluate the Anomaly Detection model you have trained but the results won't be meaningful as you have trained the model with data having no anomalies, therefore it is not trained to detect the anomalies. If the evaluation dataset also does not contain any anomalies, then there is a high probability that the scores will be really good but that doesn't ensure that the model is good and would also be able to detect anomalies in a future dataset.

Try to reduce the imbalance in the data to ensure a good model is trained that can detect both anomalies and normal data with high accuracy.