{
  "id": 331320,
  "title": "Revealing time patterns of features",
  "url": "/competitions/amex-default-prediction/discussion/331320",
  "author_name": "Pavel Vodolazov",
  "post_date": "2022-06-16T16:29:37.171000",
  "votes": 14,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Looks like vast amount of features have seasonal patterns - it may affect the model in many aspects: such as data drift and bad performance on unseen data. Also even simple feature engineering may be less accurate - if feature has stable positive trend, then max value will be closer to the last payment. We may consider normalize data on daily level to avoid those mistakes. It also may open additional dimension for feature extraction: for example we may use seasonality or trend features as additional features.</p>\n<p><img src=\"https://imgur.com/a/EaVUd0z\" alt=\"Example\"></p>\n<p>there is a <a href=\"https://www.kaggle.com/code/pavelvod/amex-eda-revealing-time-patterns-of-features\" target=\"_blank\">notebook</a> with full code</p>",
  "messages": [
    {
      "id": 1822750,
      "postDate": "2022-06-16T16:29:37.170Z",
      "content": "<p>Looks like vast amount of features have seasonal patterns - it may affect the model in many aspects: such as data drift and bad performance on unseen data. Also even simple feature engineering may be less accurate - if feature has stable positive trend, then max value will be closer to the last payment. We may consider normalize data on daily level to avoid those mistakes. It also may open additional dimension for feature extraction: for example we may use seasonality or trend features as additional features.</p>\n<p><img src=\"https://imgur.com/a/EaVUd0z\" alt=\"Example\"></p>\n<p>there is a <a href=\"https://www.kaggle.com/code/pavelvod/amex-eda-revealing-time-patterns-of-features\" target=\"_blank\">notebook</a> with full code</p>",
      "rawMarkdown": "Looks like vast amount of features have seasonal patterns - it may affect the model in many aspects: such as data drift and bad performance on unseen data. Also even simple feature engineering may be less accurate - if feature has stable positive trend, then max value will be closer to the last payment. We may consider normalize data on daily level to avoid those mistakes. It also may open additional dimension for feature extraction: for example we may use seasonality or trend features as additional features.\n\n![Example](https://imgur.com/a/EaVUd0z)\n\nthere is a [notebook](https://www.kaggle.com/code/pavelvod/amex-eda-revealing-time-patterns-of-features) with full code",
      "votes": 13
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1822750": "Looks like vast amount of features have seasonal patterns - it may affect the model in many aspects: such as data drift and bad performance on unseen data. Also even simple feature engineering may be less accurate - if feature has stable positive trend, then max value will be closer to the last payment. We may consider normalize data on daily level to avoid those mistakes. It also may open additional dimension for feature extraction: for example we may use seasonality or trend features as additional features.\n\n![Example](https://imgur.com/a/EaVUd0z)\n\nthere is a [notebook](https://www.kaggle.com/code/pavelvod/amex-eda-revealing-time-patterns-of-features) with full code"
  }
}