default class confidence is less than set by classier parameter confidence

[Sandy4321]

default class confidence is less than set by classier parameter confidence
what can be done to get default class predictions to be reliable ?

in example below default_class_confidence is 0.26 , but was used cba = CBA(confidence= 0.7)

import unittest
import pandas as pd
from pyarc import CBA
from pyarc.data_structures import (
TransactionDB
)
import os

header1 = ['F1' ,'F2','F3','F4' ,'F5','Target']

rows1 = [
[1, 1, 0, 0, 0, 1],
[1, 1, 0, 0, 0, 0],
[1, 1, 0, 0, 0, 0],
[0, 1, 1, 0, 0, 0],
[1, 1, 1, 0, 0, 1],
[1, 1, 1, 0, 0, 0],
[0, 1, 1, 0, 0, 1],
[1, 1, 1, 0, 0, 0],
[1, 1, 0, 1, 0, 1],
[1, 0, 0, 1, 1, 0],
[1, 0, 1, 1, 1, 1],
[1, 1, 0, 1, 1, 0],
[0, 1, 1, 1, 1, 0],
[1, 0, 1, 1, 0, 0],
[1, 0, 0, 1, 0, 0],
[0, 0, 0, 1, 0, 0],
[1, 0, 0, 1, 0, 0],
[0, 1, 0, 0, 0, 0],
[0, 1, 0, 0, 0, 0],
]

transactions = TransactionDB(rows1, header1)

cba = CBA(confidence= 0.7)

cba.fit(transactions)
print(cba.predict(transactions) )
[print(x) for x in cba.clf.rules]
cba.clf.rules
cba.clf.default_class
cba.clf.default_class_attribute
cba.clf.default_class_support
print('default_class_confidence' , cba.clf.default_class_confidence)
print('\n *** predict_matched_rules ***')
[print(x) for x in cba.predict_matched_rules(transactions) ]

print('\n predict_probability')
print([int(x * 100) for x in cba.predict_probability(transactions) ])
q=0

then output
['0', '0', '0', '0', '0', '0', '0', '0', '0', '0', '1', '0', '0', '0', '0', '0', '0', '0', '0']
CAR {F3=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 36
CAR {F5=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 39
CAR {F3=0,F1=0} => {Target=0} sup: 0.16 conf: 1.00 len: 3, id: 20
CAR {F5=1,F3=0} => {Target=0} sup: 0.11 conf: 1.00 len: 3, id: 5
CAR {F5=1,F2=1} => {Target=0} sup: 0.11 conf: 1.00 len: 3, id: 6
CAR {F1=0} => {Target=0} sup: 0.26 conf: 0.83 len: 2, id: 24
CAR {F3=0} => {Target=0} sup: 0.47 conf: 0.82 len: 2, id: 60
CAR {F1=1,F5=0,F3=1} => {Target=0} sup: 0.16 conf: 0.75 len: 4, id: 41
default_class_confidence 0.2631578947368421 !!!!!!!!!!!!!!!

*** predict_matched_rules ***
CAR {F3=0} => {Target=0} sup: 0.47 conf: 0.82 len: 2, id: 60
CAR {F3=0} => {Target=0} sup: 0.47 conf: 0.82 len: 2, id: 60
CAR {F3=0} => {Target=0} sup: 0.47 conf: 0.82 len: 2, id: 60
CAR {F1=0} => {Target=0} sup: 0.26 conf: 0.83 len: 2, id: 24
CAR {F1=1,F5=0,F3=1} => {Target=0} sup: 0.16 conf: 0.75 len: 4, id: 41
CAR {F1=1,F5=0,F3=1} => {Target=0} sup: 0.16 conf: 0.75 len: 4, id: 41
CAR {F1=0} => {Target=0} sup: 0.26 conf: 0.83 len: 2, id: 24
CAR {F1=1,F5=0,F3=1} => {Target=0} sup: 0.16 conf: 0.75 len: 4, id: 41
CAR {F3=0} => {Target=0} sup: 0.47 conf: 0.82 len: 2, id: 60
CAR {F3=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 36
CAR {} => {Target=1} sup: 0.26 conf: 0.26 len: 1, id: 64
CAR {F5=1,F3=0} => {Target=0} sup: 0.11 conf: 1.00 len: 3, id: 5
CAR {F5=1,F2=1} => {Target=0} sup: 0.11 conf: 1.00 len: 3, id: 6
CAR {F5=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 39
CAR {F3=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 36
CAR {F3=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 36
CAR {F3=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 36
CAR {F3=0,F1=0} => {Target=0} sup: 0.16 conf: 1.00 len: 3, id: 20
CAR {F3=0,F1=0} => {Target=0} sup: 0.16 conf: 1.00 len: 3, id: 20

predict_probability
[81, 81, 81, 83, 75, 75, 83, 75, 81, 100, 26, 100, 100, 100, 100, 100, 100, 100, 100]
^
???!!!!

[Sandy4321]

another example for this problem
it would be nice to share precision between both classes 0 and 1

but not have one of them really bad and another really good

#S_8_Mar_test_classifier_N7.py
#S_8_Mar_test_classifier_N6.py
#S_8_Mar_test_classifier_N5.py
#S_8_Mar_test_classifier_N4.py
#S_8_Mar_test_classifier.py
import unittest
import pandas as pd
from pyarc import CBA
from pyarc.data_structures import (
TransactionDB
)
import os

header1 = ['F1' ,'F2','F3','F4' ,'F5','Target']

if 1:

rows1 = [
[1,	1,	0,	0,	1,	1],
[1,	1,	0,	0,	0,	0],
[1,	1,	0,	0,	0,	0],
[0,	1,	1,	0,	0,	0],
[1,	1,	1,	0,	0,	1],
[1,	1,	1,	0,	0,	0],
[0,	1,	1,	0,	0,	1],
[1,	1,	1,	0,	0,	0],
[1,	1,	0,	1,	0,	1],
[1,	0,	0,	1,	1,	0],
[1,	0,	1,	1,	1,	1],
[1,	1,	0,	1,	1,	0],
[0,	1,	1,	1,	1,	0],
[1,	0,	1,	1,	0,	0],
[1,	0,	0,	1,	0,	0],
[0,	0,	0,	1,	0,	0],
[1,	0,	0,	1,	0,	0],
[0,	1,	0,	0,	0,	0],
[0,	1,	0,	0,	0,	0],
]

#rows1 = [ x[0:5] + [abs(x[5] -1)] for x in rows1]

print('data')
[ print(x[0] , ' ' , x[1]) for x in zip( rows1 , list(range(len(rows1))) ) ]

transactions = TransactionDB(rows1, header1)

cba = CBA(confidence= 0.9)

cba.fit(transactions)
print(cba.predict(transactions) )
[print(x) for x in cba.clf.rules]
cba.clf.rules
print('default_class is ' , cba.clf.default_class)
cba.clf.default_class_attribute
cba.clf.default_class_support
print('default_class_confidence' , cba.clf.default_class_confidence)
print('\n *** predict_matched_rules ***')
[print(x) for x in cba.predict_matched_rules(transactions) ]

print('\n predict_probability')
print([int(x * 100) for x in cba.predict_probability(transactions) ])
q=0

---

data
[1, 1, 0, 0, 1, 1] 0
[1, 1, 0, 0, 0, 0] 1
[1, 1, 0, 0, 0, 0] 2
[0, 1, 1, 0, 0, 0] 3
[1, 1, 1, 0, 0, 1] 4
[1, 1, 1, 0, 0, 0] 5
[0, 1, 1, 0, 0, 1] 6
[1, 1, 1, 0, 0, 0] 7
[1, 1, 0, 1, 0, 1] 8
[1, 0, 0, 1, 1, 0] 9
[1, 0, 1, 1, 1, 1] 10
[1, 1, 0, 1, 1, 0] 11
[0, 1, 1, 1, 1, 0] 12
[1, 0, 1, 1, 0, 0] 13
[1, 0, 0, 1, 0, 0] 14
[0, 0, 0, 1, 0, 0] 15
[1, 0, 0, 1, 0, 0] 16
[0, 1, 0, 0, 0, 0] 17
[0, 1, 0, 0, 0, 0] 18
['1', '0', '0', '1', '1', '1', '1', '1', '1', '0', '1', '0', '0', '0', '0', '0', '0', '0', '0']
CAR {F3=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 21
CAR {F2=0,F5=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 23
CAR {F5=0,F3=0,F4=0} => {Target=0} sup: 0.21 conf: 1.00 len: 4, id: 27
CAR {F4=1,F1=0} => {Target=0} sup: 0.11 conf: 1.00 len: 3, id: 3
CAR {F4=1,F3=0,F5=1} => {Target=0} sup: 0.11 conf: 1.00 len: 4, id: 1
default_class is 1
default_class_confidence 0.2631578947368421

*** predict_matched_rules ***
CAR {} => {Target=1} sup: 0.26 conf: 0.26 len: 1, id: 28
CAR {F5=0,F3=0,F4=0} => {Target=0} sup: 0.21 conf: 1.00 len: 4, id: 27
CAR {F5=0,F3=0,F4=0} => {Target=0} sup: 0.21 conf: 1.00 len: 4, id: 27
CAR {} => {Target=1} sup: 0.26 conf: 0.26 len: 1, id: 29
CAR {} => {Target=1} sup: 0.26 conf: 0.26 len: 1, id: 30
CAR {} => {Target=1} sup: 0.26 conf: 0.26 len: 1, id: 31
CAR {} => {Target=1} sup: 0.26 conf: 0.26 len: 1, id: 32
CAR {} => {Target=1} sup: 0.26 conf: 0.26 len: 1, id: 33
CAR {} => {Target=1} sup: 0.26 conf: 0.26 len: 1, id: 34
CAR {F3=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 21
CAR {} => {Target=1} sup: 0.26 conf: 0.26 len: 1, id: 35
CAR {F4=1,F3=0,F5=1} => {Target=0} sup: 0.11 conf: 1.00 len: 4, id: 1
CAR {F4=1,F1=0} => {Target=0} sup: 0.11 conf: 1.00 len: 3, id: 3
CAR {F2=0,F5=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 23
CAR {F3=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 21
CAR {F3=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 21
CAR {F3=0,F2=0} => {Target=0} sup: 0.21 conf: 1.00 len: 3, id: 21
CAR {F5=0,F3=0,F4=0} => {Target=0} sup: 0.21 conf: 1.00 len: 4, id: 27
CAR {F5=0,F3=0,F4=0} => {Target=0} sup: 0.21 conf: 1.00 len: 4, id: 27

predict_probability
[26, 100, 100, 26, 26, 26, 26, 26, 26, 100, 26, 100, 100, 100, 100, 100, 100, 100, 100]

[jirifilip]

Hello Sandy,

the default class confidence issue is a bit complicated. If you look at the original article by Liu et al., you find that default class is calculated from the remaining data that the rules in the classifier were not able to classify.

We then look at this remaining minority of data and calculate the most frequent class value. So, given the remaining data, the default class is the best prediction we can make. But, for the default rule to function as a rule, we must calculate the real confidence and support for it. This real confidence and support can be less than the threshold that we specified. It does not matter because
when applying the rules it the classifier, only the "minority" of data will get to the default rule. And for the minority, the default class is the best possible prediction.

I recommend looking at the article. But if you have any further question, I will do my best to answer.