{
  "id": 262751,
  "title": "Prediction Accuracy vs Area under ROC",
  "url": "/competitions/rsna-miccai-brain-tumor-radiogenomic-classification/discussion/262751",
  "author_name": "",
  "post_date": "2021-08-07T13:11:07.508536Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I've gone over some articles and realised the two metrics are disparate.<br>\nCan someone help me get some intuitive insight to how the two are different and what to look out for. Maybe provide links to resources that explain area under ROC.</p>\n<p>More specifically i am able to accurately predict the class with 80% accuracy but ending up with AUC under ROC &lt; 0.600. Really struggling in understanding how to improve my score.</p>\n<p>Thanks</p>",
  "messages": [
    {
      "id": "1457538",
      "postDate": "08/07/2021 13:11:07",
      "content": "<p>I've gone over some articles and realised the two metrics are disparate.<br>\nCan someone help me get some intuitive insight to how the two are different and what to look out for. Maybe provide links to resources that explain area under ROC.</p>\n<p>More specifically i am able to accurately predict the class with 80% accuracy but ending up with AUC under ROC &lt; 0.600. Really struggling in understanding how to improve my score.</p>\n<p>Thanks</p>",
      "rawMarkdown": "I've gone over some articles and realised the two metrics are disparate.\nCan someone help me get some intuitive insight to how the two are different and what to look out for. Maybe provide links to resources that explain area under ROC.\n\nMore specifically i am able to accurately predict the class with 80% accuracy but ending up with AUC under ROC < 0.600. Really struggling in understanding how to improve my score.\n\nThanks",
      "votes": null
    },
    {
      "id": "1457794",
      "postDate": "08/07/2021 15:06:59",
      "content": "<p>One reason for the low score could be because the validation set is quite small. Which might give some unstable results which is also why the leaderboard isn't a good measure of model accuracy. If you want to learn more about auc and roc I'd recommend this youtube video <a href=\"https://youtu.be/4jRBRDbJemM\" target=\"_blank\">https://youtu.be/4jRBRDbJemM</a></p>",
      "rawMarkdown": "One reason for the low score could be because the validation set is quite small. Which might give some unstable results which is also why the leaderboard isn't a good measure of model accuracy. If you want to learn more about auc and roc I'd recommend this youtube video https://youtu.be/4jRBRDbJemM",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1457794,
      "author_name": "aristotle609",
      "author_url": "",
      "post_date": "08/07/2021 15:06:59",
      "content": "<p>One reason for the low score could be because the validation set is quite small. Which might give some unstable results which is also why the leaderboard isn't a good measure of model accuracy. If you want to learn more about auc and roc I'd recommend this youtube video <a href=\"https://youtu.be/4jRBRDbJemM\" target=\"_blank\">https://youtu.be/4jRBRDbJemM</a></p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1457538": "I've gone over some articles and realised the two metrics are disparate.\nCan someone help me get some intuitive insight to how the two are different and what to look out for. Maybe provide links to resources that explain area under ROC.\n\nMore specifically i am able to accurately predict the class with 80% accuracy but ending up with AUC under ROC < 0.600. Really struggling in understanding how to improve my score.\n\nThanks",
    "1457794": "One reason for the low score could be because the validation set is quite small. Which might give some unstable results which is also why the leaderboard isn't a good measure of model accuracy. If you want to learn more about auc and roc I'd recommend this youtube video https://youtu.be/4jRBRDbJemM"
  },
  "source": "meta"
}