{ "id": 585901, "title": "Is the demographic feature actually useful?", "url": "/competitions/cmi-detect-behavior-with-sensor-data/discussion/585901", "author_name": "", "post_date": "2025-06-24T04:22:20.903850200Z", "votes": 1, "comment_count": 9, "views": 0, "content": "

I tried to remove the demographic user and both the CV and LB score improved. Wonder is it actually useful for any of the models?

", "messages": [ { "id": "3231194", "postDate": "06/24/2025 04:22:20", "content": "

I tried to remove the demographic user and both the CV and LB score improved. Wonder is it actually useful for any of the models?

", "rawMarkdown": "I tried to remove the demographic user and both the CV and LB score improved. Wonder is it actually useful for any of the models?", "votes": null }, { "id": "3231419", "postDate": "06/24/2025 10:10:22", "content": "

Same for me. It seems removing the demographic features is improving the score. I wonder why?

", "rawMarkdown": "Same for me. It seems removing the demographic features is improving the score. I wonder why?", "votes": null }, { "id": "3231478", "postDate": "06/24/2025 12:38:02", "content": "

I tried running public models and incorporating demographics features it doesn't improve the score for me .

", "rawMarkdown": "I tried running public models and incorporating demographics features it doesn't improve the score for me .", "votes": null }, { "id": "3231481", "postDate": "06/24/2025 12:42:28", "content": "

I was thinking are the subjects all curated by CMI, so the features like gender, arm length does correlate to the target. There are not like female or male are more prone to any sort of behavior.

", "rawMarkdown": "I was thinking are the subjects all curated by CMI, so the features like gender, arm length does correlate to the target. There are not like female or male are more prone to any sort of behavior.", "votes": null }, { "id": "3231501", "postDate": "06/24/2025 13:39:17", "content": "
\n

In the study, participants performed series of repeated gestures while wearing the Helios device:

\n \n
\n

I think since participants provided data by performing movements according to directives, user-specific characteristics don't effectively help us learn personal habits, behaviors etc. It seems that as long as you follow a directive, who you are doesn't matter much. 😅 I was also hopeful about measurements like arm length though, but they didn't turn out to be very useful.

", "rawMarkdown": ">In the study, participants performed series of repeated gestures while wearing the Helios device:\n- They began a transition from “rest” position and moved their hand to the appropriate location (Transition);\n- They followed this with a short pause wherein they did nothing (Pause); and\n- Finally they performed a gesture from either the BFRB-like or non-BFRB-like category of movements (Gesture; see Table below).\n\nI think since participants provided data by performing movements according to directives, user-specific characteristics don't effectively help us learn personal habits, behaviors etc. It seems that as long as you follow a directive, who you are doesn't matter much. 😅 I was also hopeful about measurements like arm length though, but they didn't turn out to be very useful.", "votes": null }, { "id": "3231517", "postDate": "06/24/2025 14:13:05", "content": "

Hey Anil, thanks for sharing. Same here, I tried incorporating them into my IMU only model but it did not help but decrease performance slightly…

", "rawMarkdown": "Hey Anil, thanks for sharing. Same here, I tried incorporating them into my IMU only model but it did not help but decrease performance slightly...", "votes": null }, { "id": "3231548", "postDate": "06/24/2025 15:03:14", "content": "

Same here. Demographic feature drop my lb by about 0.07

", "rawMarkdown": "Same here. Demographic feature drop my lb by about 0.07", "votes": null }, { "id": "3232564", "postDate": "06/25/2025 23:02:01", "content": "

Yes, I was able to get small improvement by leveraging the handness.
\nReflecting on acc_x and rot_x measurements from left hand users make their data distribution more equivalent to right handed users.

", "rawMarkdown": "Yes, I was able to get small improvement by leveraging the handness. \nReflecting on acc_x and rot_x measurements from left hand users make their data distribution more equivalent to right handed users.", "votes": null }, { "id": "3232617", "postDate": "06/26/2025 02:27:32", "content": "

I think some are useful,others are not useful.We need make data feature selection.

", "rawMarkdown": "I think some are useful,others are not useful.We need make data feature selection.", "votes": null }, { "id": "3232760", "postDate": "06/26/2025 06:59:22", "content": "

So, which ones are useful in your case? :)

", "rawMarkdown": "So, which ones are useful in your case? :)", "votes": null } ], "comments": [ { "id": 3231419, "author_name": "hannan781265", "author_url": "", "post_date": "06/24/2025 10:10:22", "content": "

Same for me. It seems removing the demographic features is improving the score. I wonder why?

", "votes": null, "replies": [] }, { "id": 3231478, "author_name": "sayedathar11", "author_url": "", "post_date": "06/24/2025 12:38:02", "content": "

I tried running public models and incorporating demographics features it doesn't improve the score for me .

", "votes": null, "replies": [] }, { "id": 3231481, "author_name": "zhangyue199", "author_url": "", "post_date": "06/24/2025 12:42:28", "content": "

I was thinking are the subjects all curated by CMI, so the features like gender, arm length does correlate to the target. There are not like female or male are more prone to any sort of behavior.

", "votes": null, "replies": [] }, { "id": 3231501, "author_name": "nlztrk", "author_url": "", "post_date": "06/24/2025 13:39:17", "content": "
\n

In the study, participants performed series of repeated gestures while wearing the Helios device:

\n \n
\n

I think since participants provided data by performing movements according to directives, user-specific characteristics don't effectively help us learn personal habits, behaviors etc. It seems that as long as you follow a directive, who you are doesn't matter much. 😅 I was also hopeful about measurements like arm length though, but they didn't turn out to be very useful.

", "votes": null, "replies": [ { "id": 3231517, "author_name": "janbrederecke", "author_url": "", "post_date": "06/24/2025 14:13:05", "content": "

Hey Anil, thanks for sharing. Same here, I tried incorporating them into my IMU only model but it did not help but decrease performance slightly…

", "votes": null, "replies": [] } ] }, { "id": 3231548, "author_name": "i2nfinit3y", "author_url": "", "post_date": "06/24/2025 15:03:14", "content": "

Same here. Demographic feature drop my lb by about 0.07

", "votes": null, "replies": [] }, { "id": 3232564, "author_name": "ikarosilva", "author_url": "", "post_date": "06/25/2025 23:02:01", "content": "

Yes, I was able to get small improvement by leveraging the handness.
\nReflecting on acc_x and rot_x measurements from left hand users make their data distribution more equivalent to right handed users.

", "votes": null, "replies": [] }, { "id": 3232617, "author_name": "harperrrrrrrr", "author_url": "", "post_date": "06/26/2025 02:27:32", "content": "

I think some are useful,others are not useful.We need make data feature selection.

", "votes": null, "replies": [ { "id": 3232760, "author_name": "janbrederecke", "author_url": "", "post_date": "06/26/2025 06:59:22", "content": "

So, which ones are useful in your case? :)

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3231194": "I tried to remove the demographic user and both the CV and LB score improved. Wonder is it actually useful for any of the models?", "3231419": "Same for me. It seems removing the demographic features is improving the score. I wonder why?", "3231478": "I tried running public models and incorporating demographics features it doesn't improve the score for me .", "3231481": "I was thinking are the subjects all curated by CMI, so the features like gender, arm length does correlate to the target. There are not like female or male are more prone to any sort of behavior.", "3231501": ">In the study, participants performed series of repeated gestures while wearing the Helios device:\n- They began a transition from “rest” position and moved their hand to the appropriate location (Transition);\n- They followed this with a short pause wherein they did nothing (Pause); and\n- Finally they performed a gesture from either the BFRB-like or non-BFRB-like category of movements (Gesture; see Table below).\n\nI think since participants provided data by performing movements according to directives, user-specific characteristics don't effectively help us learn personal habits, behaviors etc. It seems that as long as you follow a directive, who you are doesn't matter much. 😅 I was also hopeful about measurements like arm length though, but they didn't turn out to be very useful.", "3231517": "Hey Anil, thanks for sharing. Same here, I tried incorporating them into my IMU only model but it did not help but decrease performance slightly...", "3231548": "Same here. Demographic feature drop my lb by about 0.07", "3232564": "Yes, I was able to get small improvement by leveraging the handness. \nReflecting on acc_x and rot_x measurements from left hand users make their data distribution more equivalent to right handed users.", "3232617": "I think some are useful,others are not useful.We need make data feature selection.", "3232760": "So, which ones are useful in your case? :)" }, "source": "meta" }