{
  "id": 183149,
  "title": "Mean of last three measurement prediction?",
  "url": "/competitions/osic-pulmonary-fibrosis-progression/discussion/183149",
  "author_name": "Dmitrij Kozachuk",
  "post_date": "2020-09-15T17:49:17.571000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>The most part of public notebook use FVC prediction per day, that is the hard task:</p>\n<ul>\n<li>you can predict FVC in the most of the day exactly, but three bad days might be chosen as last three measurement.</li>\n<li>if you solve this task as a time series one with linear model, than even small disturbance of the prediction line might greatly change the prediction in the each day.</li>\n</ul>\n<p>So, to avoid this problem we can predict only mean values of the last three measurement without any dependence of what day it is. Technically it will be constant prediction in the each day in your submission file. Does anybody use this approach?</p>",
  "messages": [
    {
      "id": 1011835,
      "postDate": "2020-09-15T17:49:17.570Z",
      "content": "<p>The most part of public notebook use FVC prediction per day, that is the hard task:</p>\n<ul>\n<li>you can predict FVC in the most of the day exactly, but three bad days might be chosen as last three measurement.</li>\n<li>if you solve this task as a time series one with linear model, than even small disturbance of the prediction line might greatly change the prediction in the each day.</li>\n</ul>\n<p>So, to avoid this problem we can predict only mean values of the last three measurement without any dependence of what day it is. Technically it will be constant prediction in the each day in your submission file. Does anybody use this approach?</p>",
      "rawMarkdown": "The most part of public notebook use FVC prediction per day, that is the hard task:\n- you can predict FVC in the most of the day exactly, but three bad days might be chosen as last three measurement.\n- if you solve this task as a time series one with linear model, than even small disturbance of the prediction line might greatly change the prediction in the each day.\n\nSo, to avoid this problem we can predict only mean values of the last three measurement without any dependence of what day it is. Technically it will be constant prediction in the each day in your submission file. Does anybody use this approach?",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1011835": "The most part of public notebook use FVC prediction per day, that is the hard task:\n- you can predict FVC in the most of the day exactly, but three bad days might be chosen as last three measurement.\n- if you solve this task as a time series one with linear model, than even small disturbance of the prediction line might greatly change the prediction in the each day.\n\nSo, to avoid this problem we can predict only mean values of the last three measurement without any dependence of what day it is. Technically it will be constant prediction in the each day in your submission file. Does anybody use this approach?"
  }
}