离群值检测

发布于 2025-02-07 05:54:21 字数 822 浏览 0 评论 0原文

我生成了这个示例中的离群

。

值尝试了Iforest，OCSVM和其他几个人

，以及在与那些无法重现显示出最佳结果的OCSVM之后（第三次图），我可能丢失了一些参数

检测此类异常值的最佳算法是什么？

数据集

index   day interval    device  nv_zscore   c_faulty    c_normal
0   0   0   0   -1.308137   0   1
1   0   1   0   -1.330926   0   1
2   0   2   0   -1.285348   0   1
3   0   3   0   -1.253444   0   1
4   0   4   0   -1.194192   0   1
... ... ... ... ... ... ...
67195   6   19  9   0.882879    1   0
67196   6   20  9   1.544848    1   0
67197   6   21  9   0.147239    1   0
67198   6   22  9   0.674702    1   0
67199   6   23  9   -0.306257   1   0

原文

The outliers in this example was generated by me

The graph bellow represents the readings from one device, in the same day over multiple weeks, normal values in purple, red values are outliers, which means values that stands out from the normal pattern

I have already tried iforest, ocsvm and few others

And after messing around with those i was unable to reproduce the ocsvm that showed the best result (3rd plot), i might have lost some parameters

What would be the best algorithm for detecting such outliers?

Dataset

index   day interval    device  nv_zscore   c_faulty    c_normal
0   0   0   0   -1.308137   0   1
1   0   1   0   -1.330926   0   1
2   0   2   0   -1.285348   0   1
3   0   3   0   -1.253444   0   1
4   0   4   0   -1.194192   0   1
... ... ... ... ... ... ...
67195   6   19  9   0.882879    1   0
67196   6   20  9   1.544848    1   0
67197   6   21  9   0.147239    1   0
67198   6   22  9   0.674702    1   0
67199   6   23  9   -0.306257   1   0

分享到QQ

分享到微博