{
  "id": 34230,
  "title": "Handling class imbalance",
  "url": "/competitions/intel-mobileodt-cervical-cancer-screening/discussion/34230",
  "author_name": "",
  "post_date": "2017-06-06T08:02:30.303714600Z",
  "votes": null,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Just a thread to benchmark various class imbalance handling techniques like oversampling, undersampling, ensembling across data, etc. Which one worked best for you? I have been only doing undersampling followed by ensembling across data without good results. Any hacks on the model parameters similar to scale_pos_weight in xgboost for Keras API that worked well? Thanks!</p>",
  "messages": [
    {
      "id": "189509",
      "postDate": "06/06/2017 08:02:30",
      "content": "<p>Just a thread to benchmark various class imbalance handling techniques like oversampling, undersampling, ensembling across data, etc. Which one worked best for you? I have been only doing undersampling followed by ensembling across data without good results. Any hacks on the model parameters similar to scale_pos_weight in xgboost for Keras API that worked well? Thanks!</p>",
      "rawMarkdown": "Just a thread to benchmark various class imbalance handling techniques like oversampling, undersampling, ensembling across data, etc. Which one worked best for you? I have been only doing undersampling followed by ensembling across data without good results. Any hacks on the model parameters similar to scale_pos_weight in xgboost for Keras API that worked well? Thanks!",
      "votes": null
    },
    {
      "id": "189612",
      "postDate": "06/06/2017 13:23:39",
      "content": "<p>Class imbalance is not too dramatic in this case and I ended up not using any imbalance correction techniques as I wasn't getting improved predictions. Neural networks are smart :)</p>",
      "rawMarkdown": "Class imbalance is not too dramatic in this case and I ended up not using any imbalance correction techniques as I wasn't getting improved predictions. Neural networks are smart :)",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 189612,
      "author_name": "sakvaua",
      "author_url": "",
      "post_date": "06/06/2017 13:23:39",
      "content": "<p>Class imbalance is not too dramatic in this case and I ended up not using any imbalance correction techniques as I wasn't getting improved predictions. Neural networks are smart :)</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "189509": "Just a thread to benchmark various class imbalance handling techniques like oversampling, undersampling, ensembling across data, etc. Which one worked best for you? I have been only doing undersampling followed by ensembling across data without good results. Any hacks on the model parameters similar to scale_pos_weight in xgboost for Keras API that worked well? Thanks!",
    "189612": "Class imbalance is not too dramatic in this case and I ended up not using any imbalance correction techniques as I wasn't getting improved predictions. Neural networks are smart :)"
  },
  "source": "meta"
}