{
  "id": 553105,
  "title": "Most of the features are useless",
  "url": "/competitions/child-mind-institute-problematic-internet-use/discussion/553105",
  "author_name": "",
  "post_date": "2024-12-23T18:20:54.462004100Z",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>So, as I suggested most of the features are useless and only confuse the model. I just removed the ones I thought were useless and the model turned out better (0.443 on Private score, but 0.419 on Public score)! The models with full set of features had lower Private scores, but higher Public scores.</p>\n<p>My best model with actigraphy data has:</p>\n<ul>\n<li><p>7 features for LightGBM:<br>\n           'Basic_Demos-Age', 'Basic_Demos-Sex',<br>\n            'CGAS-CGAS_Score', 'PAQ_A-PAQ_A_Total',<br>\n            'PAQ_C-PAQ_C_Total', 'SDS-SDS_Total_Raw',<br>\n            'SDS-SDS_Total_T'</p></li>\n<li><p>12 features for XGBoost and 12 features for CatBoost:<br>\n            'Basic_Demos-Enroll_Season', 'Basic_Demos-Age', 'Basic_Demos-Sex',<br>\n            'CGAS-Season', 'CGAS-CGAS_Score', 'PAQ_A-   Season',<br>\n            'PAQ_A-PAQ_A_Total', 'PAQ_C-   Season', 'PAQ_C-PAQ_C_Total',<br>\n            'SDS-Season', 'SDS-SDS_Total_Raw', 'SDS-SDS_Total_T'</p></li>\n</ul>",
  "messages": [
    {
      "id": "3079500",
      "postDate": "12/23/2024 18:20:54",
      "content": "<p>So, as I suggested most of the features are useless and only confuse the model. I just removed the ones I thought were useless and the model turned out better (0.443 on Private score, but 0.419 on Public score)! The models with full set of features had lower Private scores, but higher Public scores.</p>\n<p>My best model with actigraphy data has:</p>\n<ul>\n<li><p>7 features for LightGBM:<br>\n           'Basic_Demos-Age', 'Basic_Demos-Sex',<br>\n            'CGAS-CGAS_Score', 'PAQ_A-PAQ_A_Total',<br>\n            'PAQ_C-PAQ_C_Total', 'SDS-SDS_Total_Raw',<br>\n            'SDS-SDS_Total_T'</p></li>\n<li><p>12 features for XGBoost and 12 features for CatBoost:<br>\n            'Basic_Demos-Enroll_Season', 'Basic_Demos-Age', 'Basic_Demos-Sex',<br>\n            'CGAS-Season', 'CGAS-CGAS_Score', 'PAQ_A-   Season',<br>\n            'PAQ_A-PAQ_A_Total', 'PAQ_C-   Season', 'PAQ_C-PAQ_C_Total',<br>\n            'SDS-Season', 'SDS-SDS_Total_Raw', 'SDS-SDS_Total_T'</p></li>\n</ul>",
      "rawMarkdown": "So, as I suggested most of the features are useless and only confuse the model. I just removed the ones I thought were useless and the model turned out better (0.443 on Private score, but 0.419 on Public score)! The models with full set of features had lower Private scores, but higher Public scores.\n\nMy best model with actigraphy data has:\n\n-  7 features for LightGBM:\n               'Basic_Demos-Age', 'Basic_Demos-Sex',\n                'CGAS-CGAS_Score', 'PAQ_A-PAQ_A_Total',\n                'PAQ_C-PAQ_C_Total', 'SDS-SDS_Total_Raw',\n                'SDS-SDS_Total_T'\n\n- 12 features for XGBoost and 12 features for CatBoost:\n                'Basic_Demos-Enroll_Season', 'Basic_Demos-Age', 'Basic_Demos-Sex',\n                'CGAS-Season', 'CGAS-CGAS_Score', 'PAQ_A-   Season',\n                'PAQ_A-PAQ_A_Total', 'PAQ_C-   Season', 'PAQ_C-PAQ_C_Total',\n                'SDS-Season', 'SDS-SDS_Total_Raw', 'SDS-SDS_Total_T'",
      "votes": null
    },
    {
      "id": "3079532",
      "postDate": "12/23/2024 18:57:04",
      "content": "<p>Agreed - most of the features including the actigraphy ones were useless and only 4-5 features were of value. This assignment was a 1-day work at most <a href=\"https://www.kaggle.com/kalakagat\" target=\"_blank\">@kalakagat</a> </p>",
      "rawMarkdown": "Agreed - most of the features including the actigraphy ones were useless and only 4-5 features were of value. This assignment was a 1-day work at most @kalakagat",
      "votes": null
    },
    {
      "id": "3079847",
      "postDate": "12/24/2024 08:16:23",
      "content": "<p>Honestly, I hoped that actigraphy data could help reveal patterns of depressive symptoms (less activity) and sleep disturbances which most researchers attribute to PIU.</p>",
      "rawMarkdown": "Honestly, I hoped that actigraphy data could help reveal patterns of depressive symptoms (less activity) and sleep disturbances which most researchers attribute to PIU.",
      "votes": null
    },
    {
      "id": "3080206",
      "postDate": "12/24/2024 21:15:21",
      "content": "<p>Yes me too, but it was totally useless <a href=\"https://www.kaggle.com/kalakagat\" target=\"_blank\">@kalakagat</a> </p>",
      "rawMarkdown": "Yes me too, but it was totally useless @kalakagat",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3079532,
      "author_name": "ravi20076",
      "author_url": "",
      "post_date": "12/23/2024 18:57:04",
      "content": "<p>Agreed - most of the features including the actigraphy ones were useless and only 4-5 features were of value. This assignment was a 1-day work at most <a href=\"https://www.kaggle.com/kalakagat\" target=\"_blank\">@kalakagat</a> </p>",
      "votes": null,
      "replies": [
        {
          "id": 3079847,
          "author_name": "kalakagat",
          "author_url": "",
          "post_date": "12/24/2024 08:16:23",
          "content": "<p>Honestly, I hoped that actigraphy data could help reveal patterns of depressive symptoms (less activity) and sleep disturbances which most researchers attribute to PIU.</p>",
          "votes": null,
          "replies": [
            {
              "id": 3080206,
              "author_name": "ravi20076",
              "author_url": "",
              "post_date": "12/24/2024 21:15:21",
              "content": "<p>Yes me too, but it was totally useless <a href=\"https://www.kaggle.com/kalakagat\" target=\"_blank\">@kalakagat</a> </p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3079500": "So, as I suggested most of the features are useless and only confuse the model. I just removed the ones I thought were useless and the model turned out better (0.443 on Private score, but 0.419 on Public score)! The models with full set of features had lower Private scores, but higher Public scores.\n\nMy best model with actigraphy data has:\n\n-  7 features for LightGBM:\n               'Basic_Demos-Age', 'Basic_Demos-Sex',\n                'CGAS-CGAS_Score', 'PAQ_A-PAQ_A_Total',\n                'PAQ_C-PAQ_C_Total', 'SDS-SDS_Total_Raw',\n                'SDS-SDS_Total_T'\n\n- 12 features for XGBoost and 12 features for CatBoost:\n                'Basic_Demos-Enroll_Season', 'Basic_Demos-Age', 'Basic_Demos-Sex',\n                'CGAS-Season', 'CGAS-CGAS_Score', 'PAQ_A-   Season',\n                'PAQ_A-PAQ_A_Total', 'PAQ_C-   Season', 'PAQ_C-PAQ_C_Total',\n                'SDS-Season', 'SDS-SDS_Total_Raw', 'SDS-SDS_Total_T'",
    "3079532": "Agreed - most of the features including the actigraphy ones were useless and only 4-5 features were of value. This assignment was a 1-day work at most @kalakagat",
    "3079847": "Honestly, I hoped that actigraphy data could help reveal patterns of depressive symptoms (less activity) and sleep disturbances which most researchers attribute to PIU.",
    "3080206": "Yes me too, but it was totally useless @kalakagat"
  },
  "source": "meta"
}