{
  "id": 368952,
  "title": "Seasonality Trend Forecast",
  "url": "/competitions/applications-of-deep-learning-wustlfall-2022/discussion/368952",
  "author_name": "",
  "post_date": "2022-11-28T15:03:47.061900200Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I encountered a problem adding the trend back to the adjusted forecast sales for the submission data.</p>\n<p>After building the model using de-trended data and getting the forecast adjusted sales for 2021-10-01 to 2021-12-31, I need to multiply the adjusted sales by the 'trend' and 'seasonal' for the same period. However, since the 'trend' increases over time, I also have to forecast the 'trend' for 2021-10-01 to 2021-12-31. I have tried LSTM to forecast the trend, but the model seems so inaccurate that I got a score of around 20. </p>\n<p>Does anyone have any suggestions on how to get a relatively accurate trend forecast for the period of 2021-10-01 to 2021-12-31? Thanks!</p>",
  "messages": [
    {
      "id": "2047029",
      "postDate": "11/28/2022 15:03:47",
      "content": "<p>I encountered a problem adding the trend back to the adjusted forecast sales for the submission data.</p>\n<p>After building the model using de-trended data and getting the forecast adjusted sales for 2021-10-01 to 2021-12-31, I need to multiply the adjusted sales by the 'trend' and 'seasonal' for the same period. However, since the 'trend' increases over time, I also have to forecast the 'trend' for 2021-10-01 to 2021-12-31. I have tried LSTM to forecast the trend, but the model seems so inaccurate that I got a score of around 20. </p>\n<p>Does anyone have any suggestions on how to get a relatively accurate trend forecast for the period of 2021-10-01 to 2021-12-31? Thanks!</p>",
      "rawMarkdown": "I encountered a problem adding the trend back to the adjusted forecast sales for the submission data.\n\nAfter building the model using de-trended data and getting the forecast adjusted sales for 2021-10-01 to 2021-12-31, I need to multiply the adjusted sales by the 'trend' and 'seasonal' for the same period. However, since the 'trend' increases over time, I also have to forecast the 'trend' for 2021-10-01 to 2021-12-31. I have tried LSTM to forecast the trend, but the model seems so inaccurate that I got a score of around 20. \n\nDoes anyone have any suggestions on how to get a relatively accurate trend forecast for the period of 2021-10-01 to 2021-12-31? Thanks!",
      "votes": null
    },
    {
      "id": "2047514",
      "postDate": "11/28/2022 20:25:06",
      "content": "<p>I would just use a linear regression for it.  The trend is 100% linear.</p>",
      "rawMarkdown": "I would just use a linear regression for it.  The trend is 100% linear.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2047514,
      "author_name": "jeffheaton",
      "author_url": "",
      "post_date": "11/28/2022 20:25:06",
      "content": "<p>I would just use a linear regression for it.  The trend is 100% linear.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2047029": "I encountered a problem adding the trend back to the adjusted forecast sales for the submission data.\n\nAfter building the model using de-trended data and getting the forecast adjusted sales for 2021-10-01 to 2021-12-31, I need to multiply the adjusted sales by the 'trend' and 'seasonal' for the same period. However, since the 'trend' increases over time, I also have to forecast the 'trend' for 2021-10-01 to 2021-12-31. I have tried LSTM to forecast the trend, but the model seems so inaccurate that I got a score of around 20. \n\nDoes anyone have any suggestions on how to get a relatively accurate trend forecast for the period of 2021-10-01 to 2021-12-31? Thanks!",
    "2047514": "I would just use a linear regression for it.  The trend is 100% linear."
  },
  "source": "meta"
}