{
  "id": 71069,
  "title": "Why f-1 score?",
  "url": "/competitions/quora-insincere-questions-classification/discussion/71069",
  "author_name": "",
  "post_date": "2018-11-09T21:25:26.663502500Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Can anyone explain why did we use f1-score for evaluation.</p>",
  "messages": [
    {
      "id": "418403",
      "postDate": "11/09/2018 21:25:26",
      "content": "<p>Can anyone explain why did we use f1-score for evaluation.</p>",
      "rawMarkdown": "Can anyone explain why did we use f1-score for evaluation.",
      "votes": null
    },
    {
      "id": "418466",
      "postDate": "11/09/2018 23:59:17",
      "content": "<p>Hi Kartik,\nIf you look at the number of comments flagged as insincere, you will see that they are only ~7% which means that there is a huge class imbalance. So if you look at the accuracy measure, you will not get a good estimate of model performance as even a naive classifier classifying everything as 0 will get 93% accuracy.\nTo cater to this issue, we use precision and recall metrics or F1 score which is nothing but harmonic mean of precision and recall. These metrics also take false negatives and false positives into account.\nHope this helps</p>\n\n<p>Best\nAnkur</p>",
      "rawMarkdown": "Hi Kartik,\nIf you look at the number of comments flagged as insincere, you will see that they are only ~7% which means that there is a huge class imbalance. So if you look at the accuracy measure, you will not get a good estimate of model performance as even a naive classifier classifying everything as 0 will get 93% accuracy.\nTo cater to this issue, we use precision and recall metrics or F1 score which is nothing but harmonic mean of precision and recall. These metrics also take false negatives and false positives into account.\nHope this helps\n\nBest\nAnkur",
      "votes": null
    },
    {
      "id": "418895",
      "postDate": "11/10/2018 20:23:03",
      "content": "<p>Thanks for the explanation!!</p>",
      "rawMarkdown": "Thanks for the explanation!!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 418466,
      "author_name": "ankurtomar287",
      "author_url": "",
      "post_date": "11/09/2018 23:59:17",
      "content": "<p>Hi Kartik,\nIf you look at the number of comments flagged as insincere, you will see that they are only ~7% which means that there is a huge class imbalance. So if you look at the accuracy measure, you will not get a good estimate of model performance as even a naive classifier classifying everything as 0 will get 93% accuracy.\nTo cater to this issue, we use precision and recall metrics or F1 score which is nothing but harmonic mean of precision and recall. These metrics also take false negatives and false positives into account.\nHope this helps</p>\n\n<p>Best\nAnkur</p>",
      "votes": null,
      "replies": [
        {
          "id": 418895,
          "author_name": "kbhartiya83",
          "author_url": "",
          "post_date": "11/10/2018 20:23:03",
          "content": "<p>Thanks for the explanation!!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "418403": "Can anyone explain why did we use f1-score for evaluation.",
    "418466": "Hi Kartik,\nIf you look at the number of comments flagged as insincere, you will see that they are only ~7% which means that there is a huge class imbalance. So if you look at the accuracy measure, you will not get a good estimate of model performance as even a naive classifier classifying everything as 0 will get 93% accuracy.\nTo cater to this issue, we use precision and recall metrics or F1 score which is nothing but harmonic mean of precision and recall. These metrics also take false negatives and false positives into account.\nHope this helps\n\nBest\nAnkur",
    "418895": "Thanks for the explanation!!"
  },
  "source": "meta"
}