1
00:00:00,240 --> 00:00:03,270
So first, we are creating a random forest model.

2
00:00:05,240 --> 00:00:12,980
So from a and symbol, very important random forest classifier and grading, boosting classifier.

3
00:00:16,470 --> 00:00:22,740
So what is random, first class of random footage classifier is a supervised learning algorithm, which

4
00:00:22,740 --> 00:00:24,990
is mainly used for classification problems.

5
00:00:25,940 --> 00:00:32,960
As we know, that forest is made up of trees and more trees, random forest on them creates decision

6
00:00:32,960 --> 00:00:40,640
trees on data samples and then gets the prediction and finally selects the best solution by means of

7
00:00:40,640 --> 00:00:41,140
voting.

8
00:00:41,600 --> 00:00:45,790
It creates many trees and different trees, gives different answers.

9
00:00:46,160 --> 00:00:48,880
And using voting, it's the best possible answer.

10
00:00:51,000 --> 00:00:55,470
So first we had initializing random this classifier and storing it in model.

11
00:00:56,560 --> 00:00:57,700
That's the cell.

12
00:01:02,220 --> 00:01:08,600
Then the lifting of a random forest model to extreme vitrine, let's run the self.

13
00:01:14,530 --> 00:01:21,490
So to check the parameters, you can shift it up so it will show you all the different parameters which

14
00:01:21,490 --> 00:01:22,780
are used in random forest.

15
00:01:25,820 --> 00:01:30,080
So you have initializing the model, you can check it here, you can.

16
00:01:31,480 --> 00:01:37,290
So you'll find all the different parameters so you can modify any barometer if you want.

17
00:01:39,070 --> 00:01:40,000
Let's close this.

18
00:01:41,430 --> 00:01:43,920
So we have switched over to role model an next.

19
00:01:44,930 --> 00:01:47,700
Now let's put it on our test to take.

20
00:01:49,310 --> 00:01:57,500
But the model that predicted so model is a random forest, so using that predict function, they're

21
00:01:57,500 --> 00:02:00,220
predicting it Anderson.

22
00:02:01,480 --> 00:02:07,590
Of that is good inaccuracies, converting to barometer's whitest and white predicted.

23
00:02:08,430 --> 00:02:09,390
Let's run this on.

24
00:02:11,860 --> 00:02:17,920
So here we got an accuracy of fifty eight point seven percent with random forest, classify it.

25
00:02:20,010 --> 00:02:21,220
Let's proceed further.

26
00:02:21,270 --> 00:02:23,910
Now let's create Grenadian boosting morale.

27
00:02:25,320 --> 00:02:30,600
Gradient busting is a machine learning technique for classification and regression problems, which

28
00:02:30,600 --> 00:02:35,970
produces a prediction model in the far off and symbol of the production models.

29
00:02:37,570 --> 00:02:39,130
Typically like decision trees.

30
00:02:40,970 --> 00:02:46,640
The logic behind creating the posting is simple, it works on basic assumption of linear regression.

31
00:02:47,610 --> 00:02:54,330
That is the sum of its residually zero, the residual should be spread randomly around zero.

32
00:02:55,170 --> 00:03:01,330
Now think of this residual as mistakes committed by our projector model Aalto.

33
00:03:01,890 --> 00:03:05,400
Three best models are not based on such assumptions.

34
00:03:05,610 --> 00:03:12,390
But if you think logically about this assumption, we might argue that if we are able to see some pattern

35
00:03:12,390 --> 00:03:16,770
of residuals around zero, we can leverage that pattern to fit the model.

36
00:03:18,200 --> 00:03:28,580
So the intuition behind posting algorithm is to repeatedly leverage the patterns in residuals and strengthen

37
00:03:28,580 --> 00:03:35,690
a model with weak predictions and make it better once we reach a stage that residuals do not have any

38
00:03:35,690 --> 00:03:37,820
pattern that could be model.

39
00:03:38,920 --> 00:03:42,920
We can stop modeling residuals, otherwise it may lead to the.

40
00:03:44,800 --> 00:03:51,790
Algorithmically, we're minimizing our loss function such that the best loss reaches its minimal.

41
00:03:53,380 --> 00:03:54,340
Let's proceed further.

42
00:03:55,660 --> 00:03:58,990
So here we are creating a list of the world's tallest.

43
00:03:59,410 --> 00:04:02,080
And here we are storing learning rate.

44
00:04:03,150 --> 00:04:10,050
So then we're creating a for loop for learning there in a long list, you just I.

45
00:04:12,220 --> 00:04:19,370
So here we're taking a different learning rate in our classifier, so and of course, is equal to Gregan

46
00:04:19,390 --> 00:04:26,200
boosting classifier and number of estimate, the learning related to learning rate, which is this one

47
00:04:26,530 --> 00:04:29,050
first in love, first me is this one.

48
00:04:29,050 --> 00:04:29,880
Second is this one.

49
00:04:30,460 --> 00:04:33,040
So Max is going to do Max.

50
00:04:33,040 --> 00:04:34,060
That is going to do.

51
00:04:35,400 --> 00:04:37,080
And random street is called zero.

52
00:04:38,380 --> 00:04:45,470
So you're going to see a live dogfighter, extrem come little.

53
00:04:47,530 --> 00:04:52,780
So first, very initialism the classifier, and then we are putting it on extreme well-trained.

54
00:04:53,990 --> 00:04:56,540
With different learning rates, let's say.

55
00:05:01,150 --> 00:05:03,970
So here first we're printing, learning read.

56
00:05:05,410 --> 00:05:09,400
And then we're printing, training, accuracy and validation accuracy.

57
00:05:10,480 --> 00:05:15,790
So with that format, GBE underscore selectors call extrem.

58
00:05:16,930 --> 00:05:20,260
So left score gives us accuracy.

59
00:05:21,370 --> 00:05:21,880
Anderson.

60
00:05:24,710 --> 00:05:31,010
So during training, we got an accuracy of zero point 60, which is 62 percent.

61
00:05:31,970 --> 00:05:34,430
And in addition, we got an increase of 68 percent.

62
00:05:36,810 --> 00:05:38,730
Now, let's create a moral.

63
00:05:41,900 --> 00:05:44,270
Exhaustible started in summer learning method.

64
00:05:45,670 --> 00:05:50,740
Sometimes it may not be sufficient to rely upon the results of just one machine learning model.

65
00:05:52,470 --> 00:05:59,130
So and so we're learning offers a systematic solution to combine the predictive power of multiple learners.

66
00:06:00,840 --> 00:06:06,090
The models that form the ensemble, also known as beastliness, could be either from the same learning

67
00:06:06,090 --> 00:06:13,050
algorithm or different learning algorithm, bagging and boosting all too widely used and simple learners.

68
00:06:14,020 --> 00:06:16,980
So here in this, we will be using boosting technique.

69
00:06:17,900 --> 00:06:21,800
Let's just import export classifier from Zervos.

70
00:06:22,720 --> 00:06:23,320
Lecture the.

71
00:06:25,280 --> 00:06:27,980
And then initialise Juba's classify it.

72
00:06:30,810 --> 00:06:40,710
In exuberant, you left and then so exuberant, dogfighter, too extreme and Vitron, let's run undersell.

73
00:06:46,740 --> 00:06:53,810
Now, squad is called to its CB and see if that's the extent of it.

74
00:06:54,420 --> 00:06:56,870
So this will give us the accuracy score.

75
00:06:58,280 --> 00:06:59,090
Let's bring this.

76
00:07:02,930 --> 00:07:06,500
So here we got an accuracy of sixty two point seven percent.

77
00:07:11,150 --> 00:07:18,200
So comparing to other algorithms we got most accuracy with, you know, let's try supporta machine.

78
00:07:24,440 --> 00:07:31,400
Support vector machine is a supervised machine learning model that uses classification algorithms for

79
00:07:32,040 --> 00:07:33,200
classification problems.

80
00:07:34,520 --> 00:07:36,940
Support vector machine creates a support line.

81
00:07:37,940 --> 00:07:40,610
And it creates a support line, so.

82
00:07:42,080 --> 00:07:49,490
The points above the lines are yes and points below these lines are no different in that we.

83
00:07:50,600 --> 00:07:52,160
So let's proceed further.

84
00:07:52,760 --> 00:07:55,460
So if you want to read more about this, you can click on this link.

85
00:07:56,810 --> 00:07:57,830
So first import.

86
00:07:59,250 --> 00:08:06,900
From Ashkelon, not as important as we see, so to classify them, we're initializing as we see.

87
00:08:08,170 --> 00:08:11,980
With Random Street is called a zero, and Carneal is called Alabi of.

88
00:08:13,160 --> 00:08:14,900
And a story you didn't classify it.

89
00:08:16,100 --> 00:08:22,760
Then we are fighting it on Ekstrand and Vitrine, so classified outfit, Extreme Camo Mildren.

90
00:08:24,160 --> 00:08:28,600
So the parameters of equity just shifted to.

91
00:08:30,480 --> 00:08:32,610
So we have not reported it yet.

92
00:08:32,640 --> 00:08:35,280
That's why it's not showing just shift in the.

93
00:08:37,870 --> 00:08:39,010
So they have successfully.

94
00:08:40,410 --> 00:08:49,560
We deliver a service immortal, so you shift it into all the different parameters and SBC.

95
00:08:50,840 --> 00:08:53,360
And you can also read about the new default parameter.

96
00:08:56,730 --> 00:09:00,270
Like Col's here, the default rate is above.

97
00:09:01,670 --> 00:09:04,640
And you can use linear poly sigmoid.

98
00:09:06,380 --> 00:09:07,060
Etc.

99
00:09:09,090 --> 00:09:10,650
Now, let's close this.

100
00:09:11,810 --> 00:09:14,620
Now, let's predict on our test said.

101
00:09:15,690 --> 00:09:19,470
So I predict basically those classifier that predict that.

102
00:09:20,970 --> 00:09:21,990
LZ Granderson.

103
00:09:25,240 --> 00:09:27,340
Now, let's check the accuracy of over.

104
00:09:28,290 --> 00:09:30,660
So to classify, LZ Granderson.

105
00:09:32,230 --> 00:09:34,870
So here we got an accuracy of 58 percent.

106
00:09:37,650 --> 00:09:38,130
So.

107
00:09:39,290 --> 00:09:45,110
We have implemented artificial neural networks at random for this classifier, creating the boosting,

108
00:09:46,110 --> 00:09:47,990
boosting and then support the machine.

109
00:09:50,670 --> 00:09:56,550
So out of this, we got very good accuracy with a boost and also we got a very good accuracy with artificial

110
00:09:56,550 --> 00:10:02,010
neural networks because in artificial neural networks, we have not performed any future engineering.

111
00:10:02,380 --> 00:10:04,680
Still, it was working really good.

112
00:10:09,470 --> 00:10:15,620
Even though we got a very good accuracy, but this can help the banks in knowing whether the customer

113
00:10:15,620 --> 00:10:16,810
is risky or not.

114
00:10:19,000 --> 00:10:21,060
That's it for this country.

115
00:10:21,250 --> 00:10:22,680
See you again in the next project.