{
  "id": 77557,
  "title": "Outliers in features ",
  "url": "/competitions/elo-merchant-category-recommendation/discussion/77557",
  "author_name": "",
  "post_date": "2019-01-14T09:06:07.052792300Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>How to handle outliers in features ,especially features which are skewed ? .. Deleting such data cause information loss pertaining to other parameters  and also observed that test data has outliers. Hence deletion is not a wise idea, I guess .</p>",
  "messages": [
    {
      "id": "455627",
      "postDate": "01/14/2019 09:06:07",
      "content": "<p>How to handle outliers in features ,especially features which are skewed ? .. Deleting such data cause information loss pertaining to other parameters  and also observed that test data has outliers. Hence deletion is not a wise idea, I guess .</p>",
      "rawMarkdown": "How to handle outliers in features ,especially features which are skewed ? .. Deleting such data cause information loss pertaining to other parameters  and also observed that test data has outliers. Hence deletion is not a wise idea, I guess .",
      "votes": null
    },
    {
      "id": "455648",
      "postDate": "01/14/2019 09:42:19",
      "content": "<p>Outliers are the main sources of error in this competition. The best way to reduce the error might be to <em>find</em> these outliers and not to get the more accurate model on the rest of the data. So, you can't just delete the outliers. A lot of discussions and kernels on this competition now are about how to find these outliers and how to predict them.\nLook at :</p>\n\n<p><a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/75173\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/75173</a></p>\n\n<p><a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/76950\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/76950</a></p>\n\n<p><a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77237\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77237</a></p>",
      "rawMarkdown": "Outliers are the main sources of error in this competition. The best way to reduce the error might be to *find* these outliers and not to get the more accurate model on the rest of the data. So, you can't just delete the outliers. A lot of discussions and kernels on this competition now are about how to find these outliers and how to predict them.\nLook at :\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/75173\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/76950\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77237",
      "votes": null
    },
    {
      "id": "455665",
      "postDate": "01/14/2019 10:27:48",
      "content": "<p>@Pierre-Nicolas Piquin . Thanks..</p>",
      "rawMarkdown": "Pierre-Nicolas Piquin . Thanks..",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 455648,
      "author_name": "pierrenicolaspiquin",
      "author_url": "",
      "post_date": "01/14/2019 09:42:19",
      "content": "<p>Outliers are the main sources of error in this competition. The best way to reduce the error might be to <em>find</em> these outliers and not to get the more accurate model on the rest of the data. So, you can't just delete the outliers. A lot of discussions and kernels on this competition now are about how to find these outliers and how to predict them.\nLook at :</p>\n\n<p><a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/75173\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/75173</a></p>\n\n<p><a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/76950\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/76950</a></p>\n\n<p><a href=\"https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77237\">https://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77237</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 455665,
      "author_name": "leninsolver",
      "author_url": "",
      "post_date": "01/14/2019 10:27:48",
      "content": "<p>@Pierre-Nicolas Piquin . Thanks..</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "455627": "How to handle outliers in features ,especially features which are skewed ? .. Deleting such data cause information loss pertaining to other parameters  and also observed that test data has outliers. Hence deletion is not a wise idea, I guess .",
    "455648": "Outliers are the main sources of error in this competition. The best way to reduce the error might be to *find* these outliers and not to get the more accurate model on the rest of the data. So, you can't just delete the outliers. A lot of discussions and kernels on this competition now are about how to find these outliers and how to predict them.\nLook at :\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/75173\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/76950\n\nhttps://www.kaggle.com/c/elo-merchant-category-recommendation/discussion/77237",
    "455665": "Pierre-Nicolas Piquin . Thanks.."
  },
  "source": "meta"
}