ghost
commented
6 years ago With working on Random survival Forest,the result is enhanced more than last time,but still around 83% which is still lower than yours .I think it may limit by the method .As far the feature filtering ,I have no idea for now.In my method ,VIPM tested out the overtime ,Job level and salary are most important.
获取 Outlook for iOShttps://aka.ms/o0ukef
From: Moon notifications@github.com Sent: Sunday, February 18, 2018 7:26:54 PM To: MarioBrosECS/predicting-employee-attrition-Soton-ECS Cc: wzhyzrsyj; Mention Subject: [MarioBrosECS/predicting-employee-attrition-Soton-ECS] Feature Analysis (#25)
Now we have around 33 features I have calculated their information gain while Ted has got their correlation coefficient. The result will submit as soon as possible. We can discuss, whether should we filter some unrelated features. If so, does every algorithm need that? I think decision tree doesn't need to filter features. Because it compares information gain and information gain ratio by itself. @dhiyyahhttps://github.com/dhiyyah @hoanjinanhttps://github.com/hoanjinan @Tedbanghttps://github.com/tedbang @wzhyzrsyjhttps://github.com/wzhyzrsyj @windknow123https://github.com/windknow123
― You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/MarioBrosECS/predicting-employee-attrition-Soton-ECS/issues/25, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AfgLailEEltxOD0TYYTIMPzbTKQN9M2pks5tWHl-gaJpZM4SJz2l.
Now we have around 33 features I have calculated their information gain while Ted has got their correlation coefficient. The result will submit as soon as possible. We can discuss, whether should we filter some unrelated features. If so, does every algorithm need that? I think decision tree doesn't need to filter features. Because it compares information gain and information gain ratio by itself. @dhiyyah @hoanjinan @Tedbang @wzhyzrsyj @windknow123