I think many of you already consider that P_2 is some kind of internal AMEX credit rating, indicated by its very high AUC and normal score distribution.
\nBut what if I told you there can be more discovered? I focused on one specific feature - D_39. This has caught my attention for a while and I was able to finally figure it out! This is actually target related feature! More on that you can read in my exploratory notebook:
https://www.kaggle.com/code/raddar/deanonymized-days-overdue-feat-amex
\nHope you like it :)
", "messages": [ { "id": 1829166, "postDate": "2022-06-22T11:37:17.923Z", "content": "I think many of you already consider that P_2 is some kind of internal AMEX credit rating, indicated by its very high AUC and normal score distribution.
\nBut what if I told you there can be more discovered? I focused on one specific feature - D_39. This has caught my attention for a while and I was able to finally figure it out! This is actually target related feature! More on that you can read in my exploratory notebook:
https://www.kaggle.com/code/raddar/deanonymized-days-overdue-feat-amex
\nHope you like it :)
", "rawMarkdown": "I think many of you already consider that P_2 is some kind of internal AMEX credit rating, indicated by its very high AUC and normal score distribution.\n\nBut what if I told you there can be more discovered? I focused on one specific feature - `D_39`. This has caught my attention for a while and I was able to finally figure it out! This is actually `target` related feature! More on that you can read in my exploratory notebook:\n\nhttps://www.kaggle.com/code/raddar/deanonymized-days-overdue-feat-amex\n\nHope you like it :)", "votes": 71 }, { "id": 1829174, "postDate": "2022-06-22T11:49:04.040Z", "content": "This also reminds me when I worked in credit risk industry and had to construct similar dataset. I used to put credit rating in first position and days overdue in the second position in the dataset - this must be a pattern among people who prepares such datasets :D
", "rawMarkdown": "This also reminds me when I worked in credit risk industry and had to construct similar dataset. I used to put credit rating in first position and days overdue in the second position in the dataset - this must be a pattern among people who prepares such datasets :D", "votes": 18, "replies": [ { "id": 1829231, "postDate": "2022-06-22T13:11:54.310Z", "content": "Very interesting insight!
", "rawMarkdown": "Very interesting insight!", "votes": 1 }, { "id": 1832497, "postDate": "2022-06-25T05:21:23.563Z", "content": "I also work in the financial Industry but I wonder, we try to predict default prob, but the credit score is basically that. Do you think its the Initial rating?
", "rawMarkdown": "I also work in the financial Industry but I wonder, we try to predict default prob, but the credit score is basically that. Do you think its the Initial rating?" }, { "id": 1832622, "postDate": "2022-06-25T08:08:37.147Z", "content": "I think that organizer wants to improve its internal rating (P_2).
", "rawMarkdown": "I think that organizer wants to improve its internal rating (P_2).", "votes": 5 } ] }, { "id": 1829273, "postDate": "2022-06-22T13:39:53.783Z", "content": "I got this from the first viz of @cdeotte. D_105 looks like a 'days since incident' feature, B_20 like a proportion of distressed amount.
", "rawMarkdown": "I got this from the first viz of @cdeotte. D_105 looks like a 'days since incident' feature, B_20 like a proportion of distressed amount. ", "votes": 8 }, { "id": 1832643, "postDate": "2022-06-25T08:53:42.923Z", "content": "Thank you for sharing @raddar ! Learned a lot
", "rawMarkdown": "Thank you for sharing @raddar ! Learned a lot", "votes": 1 }, { "id": 1829736, "postDate": "2022-06-22T21:57:41.010Z", "content": "you may want to check this:
\nhttps://www.kaggle.com/datasets/pradip11/amexpert-codelab-2021
Very disturbing to see this not anonymized dataset… I wonder if he got permission to publish it
", "rawMarkdown": "Very disturbing to see this not anonymized dataset... I wonder if he got permission to publish it ", "votes": 7 } ] }, { "id": 1878823, "postDate": "2022-07-31T16:48:55.967Z", "rawMarkdown": "", "votes": 1, "isDeleted": true, "replies": [ { "id": 1879045, "postDate": "2022-07-31T20:23:03.960Z", "content": "I have worked in credit scoring before. the shape of distribution and target rate indicates this. And it has the largest AUC score of all the features - which itself indicates that it is an aggregate of other features (aggregate=credit risk model in this case)
", "rawMarkdown": "I have worked in credit scoring before. the shape of distribution and target rate indicates this. And it has the largest AUC score of all the features - which itself indicates that it is an aggregate of other features (aggregate=credit risk model in this case)", "votes": 4 }, { "id": 1879298, "postDate": "2022-08-01T01:32:43.413Z", "rawMarkdown": "", "isDeleted": true } ] } ], "comments": [ { "id": 1829174, "author_name": "raddar", "author_url": "", "post_date": "2022-06-22T11:49:04.040000", "content": "This also reminds me when I worked in credit risk industry and had to construct similar dataset. I used to put credit rating in first position and days overdue in the second position in the dataset - this must be a pattern among people who prepares such datasets :D
", "votes": 18, "replies": [ { "id": 1829231, "author_name": "Ali Abdin", "author_url": "", "post_date": "2022-06-22T13:11:54.310000", "content": "Very interesting insight!
", "votes": 1, "replies": [] }, { "id": 1832497, "author_name": "Art Vandelay", "author_url": "", "post_date": "2022-06-25T05:21:23.563000", "content": "I also work in the financial Industry but I wonder, we try to predict default prob, but the credit score is basically that. Do you think its the Initial rating?
", "votes": 0, "replies": [] }, { "id": 1832622, "author_name": "raddar", "author_url": "", "post_date": "2022-06-25T08:08:37.147000", "content": "I think that organizer wants to improve its internal rating (P_2).
", "votes": 5, "replies": [] } ] }, { "id": 1829273, "author_name": "Lucas Morin", "author_url": "", "post_date": "2022-06-22T13:39:53.783000", "content": "I got this from the first viz of @cdeotte. D_105 looks like a 'days since incident' feature, B_20 like a proportion of distressed amount.
", "votes": 8, "replies": [] }, { "id": 1832643, "author_name": "Making TARS", "author_url": "", "post_date": "2022-06-25T08:53:42.923000", "content": "Thank you for sharing @raddar ! Learned a lot
", "votes": 1, "replies": [] }, { "id": 1829736, "author_name": "hengck23", "author_url": "", "post_date": "2022-06-22T21:57:41.010000", "content": "you may want to check this:
\nhttps://www.kaggle.com/datasets/pradip11/amexpert-codelab-2021
Very disturbing to see this not anonymized dataset… I wonder if he got permission to publish it
", "votes": 7, "replies": [] } ] }, { "id": 1878823, "author_name": "", "author_url": "", "post_date": "2022-07-31T16:48:55.967000", "content": "", "votes": 1, "replies": [ { "id": 1879045, "author_name": "raddar", "author_url": "", "post_date": "2022-07-31T20:23:03.960000", "content": "I have worked in credit scoring before. the shape of distribution and target rate indicates this. And it has the largest AUC score of all the features - which itself indicates that it is an aggregate of other features (aggregate=credit risk model in this case)
", "votes": 4, "replies": [] }, { "id": 1879298, "author_name": "", "author_url": "", "post_date": "2022-08-01T01:32:43.413000", "content": "", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "1829166": "I think many of you already consider that P_2 is some kind of internal AMEX credit rating, indicated by its very high AUC and normal score distribution.\n\nBut what if I told you there can be more discovered? I focused on one specific feature - `D_39`. This has caught my attention for a while and I was able to finally figure it out! This is actually `target` related feature! More on that you can read in my exploratory notebook:\n\nhttps://www.kaggle.com/code/raddar/deanonymized-days-overdue-feat-amex\n\nHope you like it :)", "1829174": "This also reminds me when I worked in credit risk industry and had to construct similar dataset. I used to put credit rating in first position and days overdue in the second position in the dataset - this must be a pattern among people who prepares such datasets :D", "1829273": "I got this from the first viz of @cdeotte. D_105 looks like a 'days since incident' feature, B_20 like a proportion of distressed amount. ", "1832643": "Thank you for sharing @raddar ! Learned a lot", "1829736": "you may want to check this:\nhttps://www.kaggle.com/datasets/pradip11/amexpert-codelab-2021", "1878823": "" } }