{
  "id": 85093,
  "title": "Predicting when time to earthquake is long",
  "url": "/competitions/LANL-Earthquake-Prediction/discussion/85093",
  "author_name": "",
  "post_date": "2019-03-21T17:39:37.301670Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Has anyone had any luck in getting good performance when the time to quake is long? (~8-9+ seconds). Every model I try seems to perform extremely poorly in this area.</p>",
  "messages": [
    {
      "id": "495931",
      "postDate": "03/21/2019 17:39:37",
      "content": "<p>Has anyone had any luck in getting good performance when the time to quake is long? (~8-9+ seconds). Every model I try seems to perform extremely poorly in this area.</p>",
      "rawMarkdown": "Has anyone had any luck in getting good performance when the time to quake is long? (~8-9+ seconds). Every model I try seems to perform extremely poorly in this area.",
      "votes": null
    },
    {
      "id": "496867",
      "postDate": "03/22/2019 17:25:57",
      "content": "<p>It seems like regression model always overestimate extremely low values and underestimate extremely high values in order to minimize error, which is why I'm working on a classifier model to predict earthquake time jump (when time reaches 0 then jumps up to ~10), but I'm struggling to get a good AUC (around 0.9 at best) due to extreme class imbalance and lack of data (there's only 16 jumps in train). Would appreciate if anyone has better ideas.</p>",
      "rawMarkdown": "It seems like regression model always overestimate extremely low values and underestimate extremely high values in order to minimize error, which is why I'm working on a classifier model to predict earthquake time jump (when time reaches 0 then jumps up to ~10), but I'm struggling to get a good AUC (around 0.9 at best) due to extreme class imbalance and lack of data (there's only 16 jumps in train). Would appreciate if anyone has better ideas.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 496867,
      "author_name": "cjinny",
      "author_url": "",
      "post_date": "03/22/2019 17:25:57",
      "content": "<p>It seems like regression model always overestimate extremely low values and underestimate extremely high values in order to minimize error, which is why I'm working on a classifier model to predict earthquake time jump (when time reaches 0 then jumps up to ~10), but I'm struggling to get a good AUC (around 0.9 at best) due to extreme class imbalance and lack of data (there's only 16 jumps in train). Would appreciate if anyone has better ideas.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "495931": "Has anyone had any luck in getting good performance when the time to quake is long? (~8-9+ seconds). Every model I try seems to perform extremely poorly in this area.",
    "496867": "It seems like regression model always overestimate extremely low values and underestimate extremely high values in order to minimize error, which is why I'm working on a classifier model to predict earthquake time jump (when time reaches 0 then jumps up to ~10), but I'm struggling to get a good AUC (around 0.9 at best) due to extreme class imbalance and lack of data (there's only 16 jumps in train). Would appreciate if anyone has better ideas."
  },
  "source": "meta"
}