{
  "id": 389852,
  "title": "Solution with score 0.88843 (Late Submission)",
  "url": "/competitions/ga-customer-revenue-prediction/discussion/389852",
  "author_name": "Anmol Kohli",
  "post_date": "2023-02-23T07:39:21.645000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<h3>Feature extraction and selection</h3>\n<p>I am using these features - pageviews, isMobile, channelGrouping, country, metro, and newVisits_sum. The categorical variables (channelGrouping, country, and metro) are transformed using one-hot encoding and applied PCA to reduce the number of columns to 24 (roughly 10% of the total one-hot encoded columns).</p>\n<h3>Model</h3>\n<p>LinearSVR - A variant of SVR with a linear kernel. This gave the best performance compared to SVR, RandomForest, XGBoost, CatBoost, and LGM Boost.</p>\n<h3>Code</h3>\n<p><a href=\"https://www.kaggle.com/code/anmolkohli1/google-analytics-revenue-prediction\" target=\"_blank\">https://www.kaggle.com/code/anmolkohli1/google-analytics-revenue-prediction</a></p>",
  "messages": [
    {
      "id": 2156251,
      "postDate": "2023-02-23T07:39:21.647Z",
      "content": "<h3>Feature extraction and selection</h3>\n<p>I am using these features - pageviews, isMobile, channelGrouping, country, metro, and newVisits_sum. The categorical variables (channelGrouping, country, and metro) are transformed using one-hot encoding and applied PCA to reduce the number of columns to 24 (roughly 10% of the total one-hot encoded columns).</p>\n<h3>Model</h3>\n<p>LinearSVR - A variant of SVR with a linear kernel. This gave the best performance compared to SVR, RandomForest, XGBoost, CatBoost, and LGM Boost.</p>\n<h3>Code</h3>\n<p><a href=\"https://www.kaggle.com/code/anmolkohli1/google-analytics-revenue-prediction\" target=\"_blank\">https://www.kaggle.com/code/anmolkohli1/google-analytics-revenue-prediction</a></p>",
      "rawMarkdown": "### Feature extraction and selection\nI am using these features - pageviews, isMobile, channelGrouping, country, metro, and newVisits_sum. The categorical variables (channelGrouping, country, and metro) are transformed using one-hot encoding and applied PCA to reduce the number of columns to 24 (roughly 10% of the total one-hot encoded columns).\n\n### Model\nLinearSVR - A variant of SVR with a linear kernel. This gave the best performance compared to SVR, RandomForest, XGBoost, CatBoost, and LGM Boost.\n\n### Code\nhttps://www.kaggle.com/code/anmolkohli1/google-analytics-revenue-prediction",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "2156251": "### Feature extraction and selection\nI am using these features - pageviews, isMobile, channelGrouping, country, metro, and newVisits_sum. The categorical variables (channelGrouping, country, and metro) are transformed using one-hot encoding and applied PCA to reduce the number of columns to 24 (roughly 10% of the total one-hot encoded columns).\n\n### Model\nLinearSVR - A variant of SVR with a linear kernel. This gave the best performance compared to SVR, RandomForest, XGBoost, CatBoost, and LGM Boost.\n\n### Code\nhttps://www.kaggle.com/code/anmolkohli1/google-analytics-revenue-prediction"
  }
}