{
  "id": 157255,
  "title": "Weighting false positives vs false negatives",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/157255",
  "author_name": "Rambo",
  "post_date": "2020-06-09T23:25:57.103000",
  "votes": 0,
  "comment_count": 3,
  "views": 0,
  "content": "<p>The number of malignant cases is a fraction of benign examples. Getting an accuracy of 90+% is trivial by simply guessing every example as benign.\nBut that is extremely dangerous --telling people they don't have cancer when they do.\nHow are false negatives weighted?</p>",
  "messages": [
    {
      "id": 880114,
      "postDate": "2020-06-10T02:39:28.777Z",
      "content": "<p>auc is not about how many correct prediction. it's all how you ranks your prediction so it's very different from accuracy</p>",
      "rawMarkdown": "auc is not about how many correct prediction. it's all how you ranks your prediction so it's very different from accuracy"
    },
    {
      "id": 880104,
      "postDate": "2020-06-10T02:31:27.330Z",
      "content": "<p>AUC = 0.5 if you guess every example as benign. The AUC metric is not fooled that easily. (Yes, the accuracy metric is different as you say).</p>",
      "rawMarkdown": "AUC = 0.5 if you guess every example as benign. The AUC metric is not fooled that easily. (Yes, the accuracy metric is different as you say)."
    },
    {
      "id": 880020,
      "postDate": "2020-06-09T23:25:57.103Z",
      "content": "<p>The number of malignant cases is a fraction of benign examples. Getting an accuracy of 90+% is trivial by simply guessing every example as benign.\nBut that is extremely dangerous --telling people they don't have cancer when they do.\nHow are false negatives weighted?</p>",
      "rawMarkdown": "The number of malignant cases is a fraction of benign examples. Getting an accuracy of 90+% is trivial by simply guessing every example as benign.\nBut that is extremely dangerous --telling people they don't have cancer when they do.\nHow are false negatives weighted?\n\n"
    },
    {
      "id": 880200,
      "postDate": "2020-06-10T05:08:36.427Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 880114,
      "author_name": "DatNT",
      "author_url": "",
      "post_date": "2020-06-10T02:39:28.777000",
      "content": "<p>auc is not about how many correct prediction. it's all how you ranks your prediction so it's very different from accuracy</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 880104,
      "author_name": "Chris Deotte",
      "author_url": "",
      "post_date": "2020-06-10T02:31:27.330000",
      "content": "<p>AUC = 0.5 if you guess every example as benign. The AUC metric is not fooled that easily. (Yes, the accuracy metric is different as you say).</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 880200,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-06-10T05:08:36.427000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "880114": "auc is not about how many correct prediction. it's all how you ranks your prediction so it's very different from accuracy",
    "880104": "AUC = 0.5 if you guess every example as benign. The AUC metric is not fooled that easily. (Yes, the accuracy metric is different as you say).",
    "880020": "The number of malignant cases is a fraction of benign examples. Getting an accuracy of 90+% is trivial by simply guessing every example as benign.\nBut that is extremely dangerous --telling people they don't have cancer when they do.\nHow are false negatives weighted?\n\n",
    "880200": ""
  }
}