知识卡片
缺失处理要先判断原因和用途
删除或填补取决于缺失发生机制、变量用途和分析目标,不能仅按5%–10%比例定规则。均值填补会改变分布和关系,回归填补也有模型假设;缺失目标与缺失输入尤其不能机械同样处理。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 使用数据
原书出处 · 1 处
- · 查看原书依据
To deal with the missing data, the usual convention that is followed is to eliminate the total row or the instance if all the features in that instance are missing values. On the other hand, if a feature has several values missing (5 to 10 percent) then these values should be replaced with values using the following two methods: (1) replace the missing values using
参考来源
- 位置:《Machine Learning Fundamentals: Concepts, Models, and Applications》第05章《使用数据》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Machine Learning Fundamentals: Concepts, Models, and Applications/_epub-src/OEBPS/xhtml/17_chapter05.xhtml)
- 结论依据:摘录中关于缺失处理要先判断原因和用途的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:To deal with the missing data, the usual convention that is followed is to eliminate the total row or the instance if all the features in that instance are missing values. On the other hand, if a feature has several values missing (5 to 10 percent) then these values should be replaced with values using the following two methods: (1) replace the missing values using