{
  "id": 92406,
  "title": "How to predict the extreme values?",
  "url": "/competitions/LANL-Earthquake-Prediction/discussion/92406",
  "author_name": "",
  "post_date": "2019-05-16T08:14:17.112685100Z",
  "votes": 3,
  "comment_count": 1,
  "views": 0,
  "content": "<p>At this point, I think we have to agree that the main concern is to understand how to well predict the TTF on the extreme (right after failure or right before failure). \nWhat is lacking here?\n- Not good enough features?\n- Insuficcient number of special cases, i.e., (TTF above the average)?\nWhat's your suggestions to overcome this problem?</p>\n\n<p>Below, a graph showing the predicted TTF (blue) and the measured TTF (green) using lgb\n<img src=\"https://i.imgur.com/LU6NJBN.jpg\" alt=\"\"></p>",
  "messages": [
    {
      "id": "532113",
      "postDate": "05/16/2019 08:14:17",
      "content": "<p>At this point, I think we have to agree that the main concern is to understand how to well predict the TTF on the extreme (right after failure or right before failure). \nWhat is lacking here?\n- Not good enough features?\n- Insuficcient number of special cases, i.e., (TTF above the average)?\nWhat's your suggestions to overcome this problem?</p>\n\n<p>Below, a graph showing the predicted TTF (blue) and the measured TTF (green) using lgb\n<img src=\"https://i.imgur.com/LU6NJBN.jpg\" alt=\"\"></p>",
      "rawMarkdown": "At this point, I think we have to agree that the main concern is to understand how to well predict the TTF on the extreme (right after failure or right before failure). \nWhat is lacking here?\n- Not good enough features?\n- Insuficcient number of special cases, i.e., (TTF above the average)?\nWhat's your suggestions to overcome this problem?\n\nBelow, a graph showing the predicted TTF (blue) and the measured TTF (green) using lgb\n![](https://i.imgur.com/LU6NJBN.jpg)",
      "votes": null
    },
    {
      "id": "532280",
      "postDate": "05/16/2019 15:02:44",
      "content": "<p>You could follow the strategies shown in:</p>\n\n<p><a href=\"https://www.kaggle.com/daisukelab/balancing-dataset-while-earthquake-is-happening\">https://www.kaggle.com/daisukelab/balancing-dataset-while-earthquake-is-happening</a> More Samples</p>\n\n<p><a href=\"https://www.kaggle.com/zikazika/useful-new-features-and-a-optimised-model\">https://www.kaggle.com/zikazika/useful-new-features-and-a-optimised-model</a> Balancing Dataset</p>\n\n<p>But be aware that probably you wont get an LB score below 1.50</p>",
      "rawMarkdown": "You could follow the strategies shown in:\n\n[https://www.kaggle.com/daisukelab/balancing-dataset-while-earthquake-is-happening](https://www.kaggle.com/daisukelab/balancing-dataset-while-earthquake-is-happening) More Samples\n\n[https://www.kaggle.com/zikazika/useful-new-features-and-a-optimised-model](https://www.kaggle.com/zikazika/useful-new-features-and-a-optimised-model) Balancing Dataset\n\nBut be aware that probably you wont get an LB score below 1.50",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 532280,
      "author_name": "superluminal098",
      "author_url": "",
      "post_date": "05/16/2019 15:02:44",
      "content": "<p>You could follow the strategies shown in:</p>\n\n<p><a href=\"https://www.kaggle.com/daisukelab/balancing-dataset-while-earthquake-is-happening\">https://www.kaggle.com/daisukelab/balancing-dataset-while-earthquake-is-happening</a> More Samples</p>\n\n<p><a href=\"https://www.kaggle.com/zikazika/useful-new-features-and-a-optimised-model\">https://www.kaggle.com/zikazika/useful-new-features-and-a-optimised-model</a> Balancing Dataset</p>\n\n<p>But be aware that probably you wont get an LB score below 1.50</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "532113": "At this point, I think we have to agree that the main concern is to understand how to well predict the TTF on the extreme (right after failure or right before failure). \nWhat is lacking here?\n- Not good enough features?\n- Insuficcient number of special cases, i.e., (TTF above the average)?\nWhat's your suggestions to overcome this problem?\n\nBelow, a graph showing the predicted TTF (blue) and the measured TTF (green) using lgb\n![](https://i.imgur.com/LU6NJBN.jpg)",
    "532280": "You could follow the strategies shown in:\n\n[https://www.kaggle.com/daisukelab/balancing-dataset-while-earthquake-is-happening](https://www.kaggle.com/daisukelab/balancing-dataset-while-earthquake-is-happening) More Samples\n\n[https://www.kaggle.com/zikazika/useful-new-features-and-a-optimised-model](https://www.kaggle.com/zikazika/useful-new-features-and-a-optimised-model) Balancing Dataset\n\nBut be aware that probably you wont get an LB score below 1.50"
  },
  "source": "meta"
}