二元逻辑回归

Question

我需要为二元逻辑回归项目使用 SQL Server 平衡数据集。不平衡的数据大约是 10:90%。建议我如何平衡sql server中的数据？

Answer 1

这是一种方法：

select t.*
from (select t.*,
             row_number() over (partition by target order by newid()) as seqnum,
             sum(case when target = 0 then 1 else 0 end) over () as num_0,
             sum(case when target = 1 then 1 else 0 end) over () as num_1
      from t
     ) t
where (num_0 <= num_1 and seqnum <= num_0) or
      (num_1 < num_0 and seqnum <= num_1);

这会随机化每个目标值的行。它为较稀有的目标提取所有行，为较常见的目标提取相同大小的随机样本。