{
  "id": 507674,
  "title": "Raising Awareness of Problematic Aspects in the Real Task",
  "url": "/competitions/home-credit-credit-risk-model-stability/discussion/507674",
  "author_name": "",
  "post_date": "2024-05-26T20:33:14.945163900Z",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I believe that for competitions like this one, it's very useful for both the competition host and society to understand some problematic aspect that could arise. This awareness can help address potential issues more effectively.</p>\n<p>In this post, I will highlight just three examples of problematic aspects. My goal is to encourage people to raise awareness among data scientists about other issues, point out relevant work, or showcase notebooks on Kaggle that address these concerns.</p>\n<ol>\n<li><p><strong>Directly Problematic Variables</strong></p>\n<ul>\n<li>Example: The gender of a person.</li></ul></li>\n<li><p><strong>Indirectly Problematic Variables</strong></p>\n<ul>\n<li>Variables that seem harmless at first glance but are correlated with problematic attributes. These hidden correlations can introduce bias.</li></ul></li>\n<li><p><strong>Black Box Models</strong></p>\n<ul>\n<li>Even if we remove indirectly problematic variables, black box models can still pose issues. If a model’s decision-making process cannot be explained, it raises concerns about the fairness and transparency of the decisions made.</li></ul></li>\n</ol>\n<p>I encourage everyone to share their thoughts, point out issues, and highlight relevant work or notebooks on Kaggle that address these problems. </p>\n<p>I really think the competition  host will appreciate such feedbacks as all person that could deal with such<br>\ntasks.</p>",
  "messages": [
    {
      "id": "2838139",
      "postDate": "05/26/2024 20:33:14",
      "content": "<p>I believe that for competitions like this one, it's very useful for both the competition host and society to understand some problematic aspect that could arise. This awareness can help address potential issues more effectively.</p>\n<p>In this post, I will highlight just three examples of problematic aspects. My goal is to encourage people to raise awareness among data scientists about other issues, point out relevant work, or showcase notebooks on Kaggle that address these concerns.</p>\n<ol>\n<li><p><strong>Directly Problematic Variables</strong></p>\n<ul>\n<li>Example: The gender of a person.</li></ul></li>\n<li><p><strong>Indirectly Problematic Variables</strong></p>\n<ul>\n<li>Variables that seem harmless at first glance but are correlated with problematic attributes. These hidden correlations can introduce bias.</li></ul></li>\n<li><p><strong>Black Box Models</strong></p>\n<ul>\n<li>Even if we remove indirectly problematic variables, black box models can still pose issues. If a model’s decision-making process cannot be explained, it raises concerns about the fairness and transparency of the decisions made.</li></ul></li>\n</ol>\n<p>I encourage everyone to share their thoughts, point out issues, and highlight relevant work or notebooks on Kaggle that address these problems. </p>\n<p>I really think the competition  host will appreciate such feedbacks as all person that could deal with such<br>\ntasks.</p>",
      "rawMarkdown": "I believe that for competitions like this one, it's very useful for both the competition host and society to understand some problematic aspect that could arise. This awareness can help address potential issues more effectively.\n\nIn this post, I will highlight just three examples of problematic aspects. My goal is to encourage people to raise awareness among data scientists about other issues, point out relevant work, or showcase notebooks on Kaggle that address these concerns.\n\n1. **Directly Problematic Variables**\n   - Example: The gender of a person.\n\n2. **Indirectly Problematic Variables**\n   - Variables that seem harmless at first glance but are correlated with problematic attributes. These hidden correlations can introduce bias.\n\n3. **Black Box Models**\n   - Even if we remove indirectly problematic variables, black box models can still pose issues. If a model’s decision-making process cannot be explained, it raises concerns about the fairness and transparency of the decisions made.\n\nI encourage everyone to share their thoughts, point out issues, and highlight relevant work or notebooks on Kaggle that address these problems. \n\nI really think the competition  host will appreciate such feedbacks as all person that could deal with such\ntasks.",
      "votes": null
    },
    {
      "id": "2840592",
      "postDate": "05/28/2024 06:49:56",
      "content": "<p>Hello. I'm not really sure what you are trying to say. But I suppose that you mean that the predictions might be related with sexism, racism or gender inequality. If that's not your point, what do you mean then?</p>\n<p>I think there is no limitation how you can create a model. For example if you have an idea that men (clients) are more likely to default. And you build your model based on that premise and finally your hypothesis proves true. There's nothing wrong with that. If men are more likely to default, and you utilize that statistic fact, there is nothing problematic about that.</p>",
      "rawMarkdown": "Hello. I'm not really sure what you are trying to say. But I suppose that you mean that the predictions might be related with sexism, racism or gender inequality. If that's not your point, what do you mean then?\n\nI think there is no limitation how you can create a model. For example if you have an idea that men (clients) are more likely to default. And you build your model based on that premise and finally your hypothesis proves true. There's nothing wrong with that. If men are more likely to default, and you utilize that statistic fact, there is nothing problematic about that.",
      "votes": null
    },
    {
      "id": "2840639",
      "postDate": "05/28/2024 07:17:55",
      "content": "<p>Even though there is a potential correlation between a given sex and default, it does not mean there is causation (And even someone could pretend to claim the contrary there would other considerations to take into account)</p>\n<p>While you may be aware of the above, when you apply your model and make decisions based on this model, this is what could be implictly   implied.</p>\n<p>If I declare to belong to a given sex and I apply for credit, I may be rejected not because of my capability to repay but just because of my sex. This becomes discrimination and is considered problematic by a broad part of society.</p>",
      "rawMarkdown": "Even though there is a potential correlation between a given sex and default, it does not mean there is causation (And even someone could pretend to claim the contrary there would other considerations to take into account)\n\nWhile you may be aware of the above, when you apply your model and make decisions based on this model, this is what could be implictly   implied.\n\nIf I declare to belong to a given sex and I apply for credit, I may be rejected not because of my capability to repay but just because of my sex. This becomes discrimination and is considered problematic by a broad part of society.",
      "votes": null
    },
    {
      "id": "2840654",
      "postDate": "05/28/2024 07:30:54",
      "content": "<p>That's an interesting and controversial opinion.</p>\n<p>While one might think that everyone (regardless gender, age, occupation) should have the same chance to get a loan, it's not that simple.</p>\n<p>But the entire idea of making prediction model is in its essence discriminative. That's the reality. People with certain age have advantage over people with another age. People with higher income are prefered over people with lower income. </p>\n<p>It's hard to assess which criterion is discriminative. Is decision of application based on gender acceptable? Is decision of application based on income acceptable? This topic is very controversial.</p>\n<p>In the end, there is a free market economy. And financial companies are free to be selective when making contracts with clients.</p>",
      "rawMarkdown": "That's an interesting and controversial opinion.\n\nWhile one might think that everyone (regardless gender, age, occupation) should have the same chance to get a loan, it's not that simple.\n\nBut the entire idea of making prediction model is in its essence discriminative. That's the reality. People with certain age have advantage over people with another age. People with higher income are prefered over people with lower income. \n\nIt's hard to assess which criterion is discriminative. Is decision of application based on gender acceptable? Is decision of application based on income acceptable? This topic is very controversial.\n\nIn the end, there is a free market economy. And financial companies are free to be selective when making contracts with clients.",
      "votes": null
    },
    {
      "id": "2841990",
      "postDate": "05/28/2024 19:42:58",
      "content": "<p>You may be of the opinion that since of this risk cannot be eliminated, automatation of this decisions should be avoided.<br>\nYou may be of the opinion that you can mitigate this risk (without eliminating all of it) in a way that the benefit will be greater of the damages.<br>\nYou may be of the opinion that some discriminations are positive, for example you want give more credit to woman to belance some past discrimination.<br>\netc…</p>\n<p>Here my point is that we as data scientist we should make very clear all of this risks so that decision makers could make more aware choices.</p>",
      "rawMarkdown": "You may be of the opinion that since of this risk cannot be eliminated, automatation of this decisions should be avoided.\nYou may be of the opinion that you can mitigate this risk (without eliminating all of it) in a way that the benefit will be greater of the damages.\nYou may be of the opinion that some discriminations are positive, for example you want give more credit to woman to belance some past discrimination.\netc...\n \nHere my point is that we as data scientist we should make very clear all of this risks so that decision makers could make more aware choices.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2840592,
      "author_name": "matousfamera",
      "author_url": "",
      "post_date": "05/28/2024 06:49:56",
      "content": "<p>Hello. I'm not really sure what you are trying to say. But I suppose that you mean that the predictions might be related with sexism, racism or gender inequality. If that's not your point, what do you mean then?</p>\n<p>I think there is no limitation how you can create a model. For example if you have an idea that men (clients) are more likely to default. And you build your model based on that premise and finally your hypothesis proves true. There's nothing wrong with that. If men are more likely to default, and you utilize that statistic fact, there is nothing problematic about that.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2840639,
          "author_name": "cescofran",
          "author_url": "",
          "post_date": "05/28/2024 07:17:55",
          "content": "<p>Even though there is a potential correlation between a given sex and default, it does not mean there is causation (And even someone could pretend to claim the contrary there would other considerations to take into account)</p>\n<p>While you may be aware of the above, when you apply your model and make decisions based on this model, this is what could be implictly   implied.</p>\n<p>If I declare to belong to a given sex and I apply for credit, I may be rejected not because of my capability to repay but just because of my sex. This becomes discrimination and is considered problematic by a broad part of society.</p>",
          "votes": null,
          "replies": [
            {
              "id": 2840654,
              "author_name": "matousfamera",
              "author_url": "",
              "post_date": "05/28/2024 07:30:54",
              "content": "<p>That's an interesting and controversial opinion.</p>\n<p>While one might think that everyone (regardless gender, age, occupation) should have the same chance to get a loan, it's not that simple.</p>\n<p>But the entire idea of making prediction model is in its essence discriminative. That's the reality. People with certain age have advantage over people with another age. People with higher income are prefered over people with lower income. </p>\n<p>It's hard to assess which criterion is discriminative. Is decision of application based on gender acceptable? Is decision of application based on income acceptable? This topic is very controversial.</p>\n<p>In the end, there is a free market economy. And financial companies are free to be selective when making contracts with clients.</p>",
              "votes": null,
              "replies": [
                {
                  "id": 2841990,
                  "author_name": "cescofran",
                  "author_url": "",
                  "post_date": "05/28/2024 19:42:58",
                  "content": "<p>You may be of the opinion that since of this risk cannot be eliminated, automatation of this decisions should be avoided.<br>\nYou may be of the opinion that you can mitigate this risk (without eliminating all of it) in a way that the benefit will be greater of the damages.<br>\nYou may be of the opinion that some discriminations are positive, for example you want give more credit to woman to belance some past discrimination.<br>\netc…</p>\n<p>Here my point is that we as data scientist we should make very clear all of this risks so that decision makers could make more aware choices.</p>",
                  "votes": null,
                  "replies": []
                }
              ]
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2838139": "I believe that for competitions like this one, it's very useful for both the competition host and society to understand some problematic aspect that could arise. This awareness can help address potential issues more effectively.\n\nIn this post, I will highlight just three examples of problematic aspects. My goal is to encourage people to raise awareness among data scientists about other issues, point out relevant work, or showcase notebooks on Kaggle that address these concerns.\n\n1. **Directly Problematic Variables**\n   - Example: The gender of a person.\n\n2. **Indirectly Problematic Variables**\n   - Variables that seem harmless at first glance but are correlated with problematic attributes. These hidden correlations can introduce bias.\n\n3. **Black Box Models**\n   - Even if we remove indirectly problematic variables, black box models can still pose issues. If a model’s decision-making process cannot be explained, it raises concerns about the fairness and transparency of the decisions made.\n\nI encourage everyone to share their thoughts, point out issues, and highlight relevant work or notebooks on Kaggle that address these problems. \n\nI really think the competition  host will appreciate such feedbacks as all person that could deal with such\ntasks.",
    "2840592": "Hello. I'm not really sure what you are trying to say. But I suppose that you mean that the predictions might be related with sexism, racism or gender inequality. If that's not your point, what do you mean then?\n\nI think there is no limitation how you can create a model. For example if you have an idea that men (clients) are more likely to default. And you build your model based on that premise and finally your hypothesis proves true. There's nothing wrong with that. If men are more likely to default, and you utilize that statistic fact, there is nothing problematic about that.",
    "2840639": "Even though there is a potential correlation between a given sex and default, it does not mean there is causation (And even someone could pretend to claim the contrary there would other considerations to take into account)\n\nWhile you may be aware of the above, when you apply your model and make decisions based on this model, this is what could be implictly   implied.\n\nIf I declare to belong to a given sex and I apply for credit, I may be rejected not because of my capability to repay but just because of my sex. This becomes discrimination and is considered problematic by a broad part of society.",
    "2840654": "That's an interesting and controversial opinion.\n\nWhile one might think that everyone (regardless gender, age, occupation) should have the same chance to get a loan, it's not that simple.\n\nBut the entire idea of making prediction model is in its essence discriminative. That's the reality. People with certain age have advantage over people with another age. People with higher income are prefered over people with lower income. \n\nIt's hard to assess which criterion is discriminative. Is decision of application based on gender acceptable? Is decision of application based on income acceptable? This topic is very controversial.\n\nIn the end, there is a free market economy. And financial companies are free to be selective when making contracts with clients.",
    "2841990": "You may be of the opinion that since of this risk cannot be eliminated, automatation of this decisions should be avoided.\nYou may be of the opinion that you can mitigate this risk (without eliminating all of it) in a way that the benefit will be greater of the damages.\nYou may be of the opinion that some discriminations are positive, for example you want give more credit to woman to belance some past discrimination.\netc...\n \nHere my point is that we as data scientist we should make very clear all of this risks so that decision makers could make more aware choices."
  },
  "source": "meta"
}