1
00:00:01,860 --> 00:00:09,930
In this session, we're going to see univariate analysis, what is the reason I'm considering each of

2
00:00:09,930 --> 00:00:18,810
the variables one by one, I'm analyzing each other variables in isolation, like if, let's say gender

3
00:00:18,810 --> 00:00:20,730
is a variable, I'm analyzing gender.

4
00:00:21,360 --> 00:00:23,890
If income is a variable, I'm analyzing income.

5
00:00:23,910 --> 00:00:25,770
How I'm analyzing in isolation.

6
00:00:26,900 --> 00:00:27,240
Right.

7
00:00:27,530 --> 00:00:35,420
What kind of analysis I do depends on whether the data is numerical, continues variables or categorical

8
00:00:35,420 --> 00:00:38,200
gains for is a continuous variable.

9
00:00:38,210 --> 00:00:41,980
I'll be using measures of central tendency in this push.

10
00:00:42,410 --> 00:00:44,700
What is measures of central tendency in this position?

11
00:00:45,370 --> 00:00:46,710
You will see that very shortly.

12
00:00:47,120 --> 00:00:47,370
Right.

13
00:00:47,840 --> 00:00:49,400
But why are we doing this?

14
00:00:50,030 --> 00:00:51,950
We are doing this because of two reasons.

15
00:00:51,950 --> 00:00:54,320
One, I want to ensure.

16
00:00:55,480 --> 00:00:57,910
I get a better idea about my life.

17
00:00:57,940 --> 00:00:59,080
That's very obvious.

18
00:00:59,650 --> 00:01:03,080
The second reason is because of the concept of population.

19
00:01:04,450 --> 00:01:05,910
What is this population of?

20
00:01:07,830 --> 00:01:17,460
Many times in real life situations, we want to get to work on the complete historical leader, we will

21
00:01:17,460 --> 00:01:21,470
only be provided with a sample of the historical data.

22
00:01:22,470 --> 00:01:34,380
Right, it is up to us to ensure that the sample data that we have taken up for developing here is representative

23
00:01:34,380 --> 00:01:37,200
of the entire historical dataset.

24
00:01:37,980 --> 00:01:48,270
For example, if my entire historical dataset consisted of data from all the states in India or the

25
00:01:48,270 --> 00:01:55,260
US, if I take the sample data for the purpose of.

26
00:01:56,240 --> 00:02:02,220
Creating my machine learning model, I must ensure the representativeness is adhered to.

27
00:02:02,420 --> 00:02:09,230
What I mean by that is if, for example, the Midwest, you know, in the US, right in the historical

28
00:02:09,230 --> 00:02:16,120
data, contributed to 30 percent of all the insurance cases in the sample.

29
00:02:16,130 --> 00:02:20,000
Also, I must ensure that the Midwest contributes to 30 percent.

30
00:02:21,910 --> 00:02:22,250
But.

31
00:02:23,560 --> 00:02:30,760
This is something that happens in opinion polls and all right, when they do a survey as to which party

32
00:02:30,760 --> 00:02:37,690
will win, which candidate will win, they need to ensure the person who is doing the survey needs to

33
00:02:37,690 --> 00:02:41,100
ensure that the representativeness is taken care.

34
00:02:42,180 --> 00:02:48,090
Like, if I'm doing an all India survey as to which party will come to power, I need to ensure this

35
00:02:48,090 --> 00:02:52,470
representativeness in the electorate population is taken.

36
00:02:54,430 --> 00:02:54,750
Right.

37
00:02:55,270 --> 00:03:01,450
So the same concept we're doing here also because it's a very basic concept, that is when analyzing

38
00:03:01,450 --> 00:03:08,710
the variables in isolation, it will help to validate this representativeness aspect.

39
00:03:08,710 --> 00:03:15,700
That is, whatever proportion I have in the complete dataset, which is called as population, I must

40
00:03:15,700 --> 00:03:23,310
ensure the same representation level is maintained in my sample data.

41
00:03:23,320 --> 00:03:28,490
Also, I please note that, you know, you don't get to work on the complete dataset.

42
00:03:28,510 --> 00:03:29,980
You know what is called as population.

43
00:03:30,880 --> 00:03:34,030
That can be various reasons for that and what the data can be collected.

44
00:03:34,030 --> 00:03:34,300
Right.

45
00:03:35,890 --> 00:03:37,990
Only the recent data may be available.

46
00:03:38,680 --> 00:03:38,980
Right.

47
00:03:39,970 --> 00:03:44,590
Your database may have some limitations, so you may be compromising.

48
00:03:45,560 --> 00:03:51,650
So that can be various reasons, but that's a fact of life, right, if you get to work on the complete

49
00:03:51,650 --> 00:03:52,110
data set.

50
00:03:52,130 --> 00:03:55,860
Well and good, but what if you don't get to work on the complete dataset?

51
00:03:56,600 --> 00:03:57,740
Must still continue, right.

52
00:03:58,520 --> 00:04:04,280
We have a wonderful principle of statistics, which is the representation aspect in population as a

53
00:04:04,280 --> 00:04:10,910
sample, please here to ensure that the sample that you are taking for developing machine learning model.

54
00:04:11,540 --> 00:04:11,890
Right.

55
00:04:11,900 --> 00:04:17,570
Is representative of the population or the complete dataset.

56
00:04:18,170 --> 00:04:21,350
I believe in order for it.

57
00:04:22,760 --> 00:04:29,000
So let's get back to the univariate analysis in univariate analysis, if it is a continuous data for

58
00:04:29,180 --> 00:04:29,800
a number.

59
00:04:29,810 --> 00:04:30,080
Right.

60
00:04:30,530 --> 00:04:34,950
What I do is I look at what is the measure of central tendency and what is dispersion.

61
00:04:36,110 --> 00:04:41,330
Central tendency tells me, you know, I don't have a legitimate point.

62
00:04:41,990 --> 00:04:46,620
What is the value this person tells?

63
00:04:47,090 --> 00:04:47,540
Right.

64
00:04:48,110 --> 00:04:51,500
What is the level of radiation that I have in my process?

65
00:04:53,030 --> 00:05:01,010
But if let's say I take the example of Moxa painting a glass, if I say that in all the average motss,

66
00:05:02,150 --> 00:05:03,170
say 80 percent.

67
00:05:04,960 --> 00:05:11,860
But few students have scored even 95 percent or 99 percent, and few students have scored just 10 percent

68
00:05:11,860 --> 00:05:12,480
of kids.

69
00:05:13,450 --> 00:05:18,370
So this Extreme's right, eight percent and 99 percent.

70
00:05:20,370 --> 00:05:23,140
That tells me the variation that is there.

71
00:05:23,970 --> 00:05:27,420
I need to understand this also, right?

72
00:05:27,840 --> 00:05:29,850
So that's what I'll do for my country.

73
00:05:32,430 --> 00:05:34,650
So if let's say these are the numbers.

74
00:05:35,750 --> 00:05:39,730
OK, I mean, it's nothing but the arithmetic, right?

75
00:05:40,340 --> 00:05:48,030
I add all these numbers and divide by the number of entries, which is divided by five points.

76
00:05:48,560 --> 00:05:55,910
This arithmetic average median is the midpoint after arranging the data in ascending order.

77
00:05:56,930 --> 00:06:03,210
OK, please note, I need or indeed in ascending order and the midpoint is the number that counts.

78
00:06:03,490 --> 00:06:03,830
OK.

79
00:06:05,610 --> 00:06:09,570
That's my more this the frequently occurring member.

80
00:06:11,220 --> 00:06:19,230
OK, standard deviation is the extent of deviation from the average, how much I might deviating from

81
00:06:19,230 --> 00:06:20,040
the average.

82
00:06:21,520 --> 00:06:23,620
Get that's what I do in.

83
00:06:25,190 --> 00:06:30,890
Measures of central tendency and the other concepts like percentiles, they are also used, but I'm

84
00:06:30,890 --> 00:06:32,810
not covering it here.

85
00:06:33,320 --> 00:06:37,110
If you are interested, you can go ahead and understand it.

86
00:06:38,000 --> 00:06:40,940
But this is sufficient for most of the scenarios.

87
00:06:41,380 --> 00:06:49,880
OK, you will be using median more than on other concepts like the central tendency in this push and

88
00:06:49,880 --> 00:06:54,950
concepts subsequently writing the other analysis also.

89
00:06:56,960 --> 00:07:00,810
So now let's apply to some new tendency in this position to data.

90
00:07:03,150 --> 00:07:10,400
I start with the income, right, I look at what is the mean one way to understand the variation.

91
00:07:10,680 --> 00:07:17,190
OK, yes, of course, calculating the standard, I can also plot the frequency distribution.

92
00:07:19,280 --> 00:07:21,360
Like frequency distribution.

93
00:07:23,240 --> 00:07:30,470
If I put this frequency distribution, I can find that most of the income is.

94
00:07:33,350 --> 00:07:40,200
Zero to 5000 on the next is between 5000 to 20000 and beyond 20000 the.

95
00:07:41,670 --> 00:07:46,200
Few values of it, so that's the first information I get.

96
00:07:49,730 --> 00:07:58,040
If I were to apply this apply, the concept of population was a sample based on this, if you see this

97
00:07:58,040 --> 00:08:01,800
here, you know, I can convert this into some percentages.

98
00:08:01,990 --> 00:08:07,550
Majorities in 025 dozing off of that is in five thousand to twenty thousand.

99
00:08:07,550 --> 00:08:11,880
And then few entries are there greater than 20000 of.

100
00:08:12,880 --> 00:08:20,080
This is, let's say for this my sample, I need to verify if this proportion is there in my population

101
00:08:20,080 --> 00:08:24,810
or the complete dataset or if it is not there, I'm setting myself up for failure.

102
00:08:26,480 --> 00:08:29,520
Like my forecast accuracy will definitely take a minute.

103
00:08:30,590 --> 00:08:36,810
Please stop your exercise if the sample is not representative of the population.

104
00:08:36,830 --> 00:08:37,850
You must stop.

105
00:08:39,300 --> 00:08:43,140
You must first ensure that the representativeness is followed.

106
00:08:43,230 --> 00:08:45,870
That's one of the fundamental things in.

107
00:08:47,400 --> 00:08:50,130
Machine learning or in statistics?

108
00:08:51,540 --> 00:08:53,280
Machine learning is based on statistics.

109
00:08:54,000 --> 00:08:57,250
So that's the thing you will validate.

110
00:08:57,960 --> 00:09:03,930
So now that now you've got this right for categorical variables, you know, I find out what is the

111
00:09:04,680 --> 00:09:06,190
for example, in the case of gender?

112
00:09:06,810 --> 00:09:08,190
What is the percentage of male?

113
00:09:08,200 --> 00:09:09,540
What is the percentage of female?

114
00:09:09,570 --> 00:09:09,840
Right.

115
00:09:10,590 --> 00:09:12,100
What is the percentage of graduate?

116
00:09:12,150 --> 00:09:14,930
What is the percentage of of the number of graduates?

117
00:09:14,940 --> 00:09:21,630
Number of saying that whether the individual is married or not.

118
00:09:22,650 --> 00:09:25,620
I can also put this in the form of a chart.

119
00:09:26,250 --> 00:09:30,650
So all that I explain in population was the same applies here also.

120
00:09:32,150 --> 00:09:32,520
Like.

121
00:09:33,660 --> 00:09:42,480
So please ensure that, right, so I do this for both my continuance, that's for less categorical.

122
00:09:42,730 --> 00:09:50,700
If I put it in the ground, know I'm able to understand the composition of data better because the picture

123
00:09:50,700 --> 00:09:54,240
is worth a thousand words, like to always use graphs.

124
00:09:56,520 --> 00:09:56,880
Right.

125
00:09:57,210 --> 00:10:00,950
So that's what we do in univariate analysis.

126
00:10:02,110 --> 00:10:07,500
Clear and you understand why you univariate analysis is necessary on.

127
00:10:08,810 --> 00:10:16,610
OK, so having understood this, now getting to Vivarium analysis studies how one variable interacts

128
00:10:16,610 --> 00:10:22,190
with the other way, how dependent variable interacts with independently.

129
00:10:22,760 --> 00:10:26,790
So that's what we're going to see in by video analysis.

130
00:10:27,470 --> 00:10:27,960
OK.