2

我有如下混淆矩阵:

   abcdefgh <-- 分类为
1086 7 1 0 2 4 0 0 | 7 1064
8 6 0 2 2 0 | b
0 2 1091 2 3 0 1 1 | c
0 8 0 1090 1 1 0 0 | d
1 1 1 1 597 2 2 0 | e
4 2 1 0 3 1089 0 1 | f
0 2 1 3 0 0 219 0 | G
0 0 1 0 1 4 1 443 | H

现在如何从这个混淆矩阵中找到真阳性、真阴性、假阳性和假阴性值。Weka 给我的 TP Rate 和 True positive value 一样吗?

4

2 回答 2

4

您总共有 8 个课程:a, b, c, d, e, f, g, h. 因此,您将获得 8 个不同TP, FP, FN,TN数字。例如,在a类的情况下,

TP (instance belongs to a, classified as a) = 1086
FP (instance belongs to others, classified as a) = 7 + 0 + 0 + 1 + 4 + 0 + 0 = 12
FN (instance belongs to a, classified as others) = 7 + 1 + 0 + 2 + 4 + 0 + 0 = 14
TN (instance belongs to others, classified as others) = Total instance - (TP + FP + FN)

TP rate不是TP。它是RecallTP/TP+FN

于 2015-01-27T16:06:42.953 回答
0

abcdefgh <-- 分类为

1086 7 1 0 2 4 0 0 | 一个

7 1064    8    6    0    2    2    0 |    b

0    2 1091    2    3    0    1    1 |    c

0    8    0 1090    1    1    0    0 |    d

1    1    1    1  597    2    2    0 |    e

4    2    1    0    3 1089    0    1 |    f

0    2    1    3    0    0  219    0 |    g

0    0    1    0    1    4    1    0 |    h 
于 2015-06-01T08:29:03.333 回答