{
  "id": 538403,
  "title": "MNAR Values and how to handle it",
  "url": "/competitions/child-mind-institute-problematic-internet-use/discussion/538403",
  "author_name": "Richard English",
  "post_date": "2024-10-08T18:26:47.811000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Many of the features have quite a bit of missing values, as has been discussed. But it appears that some of those  missing values have meaning for a few of the features. </p>\n<p>For example if you look at 'FGC-FGC_GSND', the difference between the mean sii value of populated entries vs nan entries, you get a .404 which seems significant. There ar a few features like this. There is clearly some bias in the missing data. </p>\n<p>I can understand why this would be, but not what to do with it.<br>\nA custom imputer would make sense but without domain knowledge i'm not sure where to start.<br>\nAnyone have thoughts on this?</p>",
  "messages": [
    {
      "id": 3010178,
      "postDate": "2024-10-08T18:26:47.810Z",
      "content": "<p>Many of the features have quite a bit of missing values, as has been discussed. But it appears that some of those  missing values have meaning for a few of the features. </p>\n<p>For example if you look at 'FGC-FGC_GSND', the difference between the mean sii value of populated entries vs nan entries, you get a .404 which seems significant. There ar a few features like this. There is clearly some bias in the missing data. </p>\n<p>I can understand why this would be, but not what to do with it.<br>\nA custom imputer would make sense but without domain knowledge i'm not sure where to start.<br>\nAnyone have thoughts on this?</p>",
      "rawMarkdown": "Many of the features have quite a bit of missing values, as has been discussed. But it appears that some of those  missing values have meaning for a few of the features. \n\nFor example if you look at 'FGC-FGC_GSND', the difference between the mean sii value of populated entries vs nan entries, you get a .404 which seems significant. There ar a few features like this. There is clearly some bias in the missing data. \n\nI can understand why this would be, but not what to do with it.\nA custom imputer would make sense but without domain knowledge i'm not sure where to start.\nAnyone have thoughts on this?",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "3010178": "Many of the features have quite a bit of missing values, as has been discussed. But it appears that some of those  missing values have meaning for a few of the features. \n\nFor example if you look at 'FGC-FGC_GSND', the difference between the mean sii value of populated entries vs nan entries, you get a .404 which seems significant. There ar a few features like this. There is clearly some bias in the missing data. \n\nI can understand why this would be, but not what to do with it.\nA custom imputer would make sense but without domain knowledge i'm not sure where to start.\nAnyone have thoughts on this?"
  }
}