{
  "id": 40462,
  "title": "What are the different approaches to predict churn-out in future on a particular day?",
  "url": "/competitions/kkbox-churn-prediction-challenge/discussion/40462",
  "author_name": "Manish Sharma",
  "post_date": "2017-10-03T04:57:05.083000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p><em>Disclaimer: This topic is not exactly related to the competition, but is very important for knowing the different approaches we can apply for the churn-out problem.</em></p>\n\n<p>Sometimes we just need to predict whether a customer is going to churn out or not without taking into consideration the number of days after which he is going to churn out or not. That case is a simple classification problem. Our usual <code>XGBoost</code> and <code>RandomForest</code> will work clearly.</p>\n\n<p>However, in cases in which we need to predict the probability of churning out after say 30 days in future, that is not exactly a simple classification problem. So what are your approaches for this problem ?</p>\n\n<hr>\n\n<p>Some approaches that I know are-</p>\n\n<p>1) Survival Analysis (Though I don't know how to do that for time-varying covariates)</p>\n\n<p>2) Regress after how many days a customer is going to churn-out, and then see whether it is greater or less than the day we need to predict for.</p>\n\n<p>3) Regress all variable to the day we need  to predict for, and then use classification to say whether they will churn-out or not on that day.</p>",
  "messages": [
    {
      "id": 226808,
      "postDate": "2017-10-03T04:57:05.083Z",
      "content": "<p><em>Disclaimer: This topic is not exactly related to the competition, but is very important for knowing the different approaches we can apply for the churn-out problem.</em></p>\n\n<p>Sometimes we just need to predict whether a customer is going to churn out or not without taking into consideration the number of days after which he is going to churn out or not. That case is a simple classification problem. Our usual <code>XGBoost</code> and <code>RandomForest</code> will work clearly.</p>\n\n<p>However, in cases in which we need to predict the probability of churning out after say 30 days in future, that is not exactly a simple classification problem. So what are your approaches for this problem ?</p>\n\n<hr>\n\n<p>Some approaches that I know are-</p>\n\n<p>1) Survival Analysis (Though I don't know how to do that for time-varying covariates)</p>\n\n<p>2) Regress after how many days a customer is going to churn-out, and then see whether it is greater or less than the day we need to predict for.</p>\n\n<p>3) Regress all variable to the day we need  to predict for, and then use classification to say whether they will churn-out or not on that day.</p>",
      "rawMarkdown": "*Disclaimer: This topic is not exactly related to the competition, but is very important for knowing the different approaches we can apply for the churn-out problem.*\n\nSometimes we just need to predict whether a customer is going to churn out or not without taking into consideration the number of days after which he is going to churn out or not. That case is a simple classification problem. Our usual `XGBoost` and `RandomForest` will work clearly.\n\nHowever, in cases in which we need to predict the probability of churning out after say 30 days in future, that is not exactly a simple classification problem. So what are your approaches for this problem ?\n\n\n----------\n\n\nSome approaches that I know are-\n\n1) Survival Analysis (Though I don't know how to do that for time-varying covariates)\n\n2) Regress after how many days a customer is going to churn-out, and then see whether it is greater or less than the day we need to predict for.\n\n3) Regress all variable to the day we need  to predict for, and then use classification to say whether they will churn-out or not on that day.",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "226808": "*Disclaimer: This topic is not exactly related to the competition, but is very important for knowing the different approaches we can apply for the churn-out problem.*\n\nSometimes we just need to predict whether a customer is going to churn out or not without taking into consideration the number of days after which he is going to churn out or not. That case is a simple classification problem. Our usual `XGBoost` and `RandomForest` will work clearly.\n\nHowever, in cases in which we need to predict the probability of churning out after say 30 days in future, that is not exactly a simple classification problem. So what are your approaches for this problem ?\n\n\n----------\n\n\nSome approaches that I know are-\n\n1) Survival Analysis (Though I don't know how to do that for time-varying covariates)\n\n2) Regress after how many days a customer is going to churn-out, and then see whether it is greater or less than the day we need to predict for.\n\n3) Regress all variable to the day we need  to predict for, and then use classification to say whether they will churn-out or not on that day."
  }
}