Can anyone help me provide more details about area under the ROC curve ? and how to choose the best curve ?
", "messages": [ { "id": "1512813", "postDate": "09/14/2021 15:45:44", "content": "Can anyone help me provide more details about area under the ROC curve ? and how to choose the best curve ?
", "rawMarkdown": "Can anyone help me provide more details about area under the ROC curve ? and how to choose the best curve ?", "votes": null }, { "id": "1513643", "postDate": "09/15/2021 09:46:38", "content": "ROC stands for Receiver Characteristic Operator, which is a name used in the 1940s for the method to evaluate detection of Japanese aircrafts from the radar signals. AUC-ROC is a curve of True Positive Rate against False Positive Rate.
\nThe best curve is simply the one that has the biggest area under it. A value of this area = 0.5 means that the classifier is random, and the higher value above this threshold, the better the classifier is in its predictions. What matters the most here is class separability, so if there is a threshold for the returned probabilities over which all positives are true, and below which all negatives are true, this is an ideal model. There are various implementations of AUC-ROC, available for example in sklearn, pytorch, or tensorflow.
@abdelkrimbeg Hello. You can check this discussion, \"AUC ROC curve\" for Beginners. I have compiled all the resources. I think it will be helpful for you. If you find it beneficial, please consider giving an upvote.
", "rawMarkdown": "abdelkrimbeg Hello. You can check this discussion, [\"AUC ROC curve\" for Beginners](https://www.kaggle.com/c/tabular-playground-series-sep-2021/discussion/269985). I have compiled all the resources. I think it will be helpful for you. If you find it beneficial, please consider giving an upvote.", "votes": null } ], "comments": [ { "id": 1513643, "author_name": "mikecho", "author_url": "", "post_date": "09/15/2021 09:46:38", "content": "ROC stands for Receiver Characteristic Operator, which is a name used in the 1940s for the method to evaluate detection of Japanese aircrafts from the radar signals. AUC-ROC is a curve of True Positive Rate against False Positive Rate.
\nThe best curve is simply the one that has the biggest area under it. A value of this area = 0.5 means that the classifier is random, and the higher value above this threshold, the better the classifier is in its predictions. What matters the most here is class separability, so if there is a threshold for the returned probabilities over which all positives are true, and below which all negatives are true, this is an ideal model. There are various implementations of AUC-ROC, available for example in sklearn, pytorch, or tensorflow.
@abdelkrimbeg Hello. You can check this discussion, \"AUC ROC curve\" for Beginners. I have compiled all the resources. I think it will be helpful for you. If you find it beneficial, please consider giving an upvote.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1512813": "Can anyone help me provide more details about area under the ROC curve ? and how to choose the best curve ?", "1513643": "ROC stands for Receiver Characteristic Operator, which is a name used in the 1940s for the method to evaluate detection of Japanese aircrafts from the radar signals. AUC-ROC is a curve of True Positive Rate against False Positive Rate.\nThe best curve is simply the one that has the biggest area under it. A value of this area = 0.5 means that the classifier is random, and the higher value above this threshold, the better the classifier is in its predictions. What matters the most here is class separability, so if there is a threshold for the returned probabilities over which all positives are true, and below which all negatives are true, this is an ideal model. There are various implementations of AUC-ROC, available for example in sklearn, pytorch, or tensorflow.", "1516080": "abdelkrimbeg Hello. You can check this discussion, [\"AUC ROC curve\" for Beginners](https://www.kaggle.com/c/tabular-playground-series-sep-2021/discussion/269985). I have compiled all the resources. I think it will be helpful for you. If you find it beneficial, please consider giving an upvote." }, "source": "meta" }