1
00:00:01,780 --> 00:00:08,590
Now that we have looked at univariate analysis and by vinyard analysis, it is time to sit back and

2
00:00:08,590 --> 00:00:10,990
again think about these videos that we have.

3
00:00:12,910 --> 00:00:16,630
Remember that we are trying to keep only relevant variables in the analysis.

4
00:00:17,970 --> 00:00:21,760
So variables like bus terminal, which has only value.

5
00:00:21,790 --> 00:00:22,330
Yes.

6
00:00:23,160 --> 00:00:24,330
To our needed I it.

7
00:00:26,110 --> 00:00:31,150
It is pretty indicative that this variable is not going to add any information to our model.

8
00:00:33,420 --> 00:00:34,250
It may be relevant.

9
00:00:34,980 --> 00:00:39,410
Maybe if a city does not have a bus terminal, the prices will be lower there.

10
00:00:40,380 --> 00:00:45,870
But we cannot see that by looking at the sample because dissemble has no observation.

11
00:00:46,110 --> 00:00:47,730
Devalued bus terminal is no.

12
00:00:48,660 --> 00:00:55,980
So since this variable is not useful, we will remove it from our day, does it?

13
00:00:57,150 --> 00:01:04,230
Think about all these points again, these four points to decide if the variables we have are relevant

14
00:01:04,230 --> 00:01:04,620
or not.

15
00:01:06,030 --> 00:01:10,530
Fosters rebuild with single unique value such as bus terminal.

16
00:01:11,340 --> 00:01:12,170
Those should be removed.

17
00:01:14,000 --> 00:01:17,990
Even for non categorical variables, if they have only one value to old.

18
00:01:18,920 --> 00:01:20,480
That is basically not a variable.

19
00:01:20,870 --> 00:01:22,700
That is basically behaving as a constant.

20
00:01:23,030 --> 00:01:25,730
We don't need such variables and we will be deleting them.

21
00:01:27,580 --> 00:01:31,450
Variables with low filtrate are the second point.

22
00:01:31,660 --> 00:01:32,410
We saw that.

23
00:01:33,280 --> 00:01:36,490
And Horsburgh Variable had eight values missing.

24
00:01:36,850 --> 00:01:41,390
So we decided that we will replace these values by the mean of the other values.

25
00:01:42,820 --> 00:01:48,640
But suppose we had only 50 values and the rest 450 values were empty.

26
00:01:49,890 --> 00:01:52,960
Does it make sense to impute values in that case?

27
00:01:54,530 --> 00:01:59,820
In such a case will not be able to capture the actual effect of that variable on the airport.

28
00:02:01,320 --> 00:02:08,910
Even if you keep the variable and impute mean value in the missing places and on the analysis nearly

29
00:02:09,000 --> 00:02:14,100
always, there will not be any significant relationship between such a variable and the output.

30
00:02:16,040 --> 00:02:18,830
So we have the option of deleting Saitoti, able to.

31
00:02:20,290 --> 00:02:20,830
Thirdly.

32
00:02:21,950 --> 00:02:25,640
Remember that businesses are working within a regulatory framework.

33
00:02:26,870 --> 00:02:30,560
I'm regulatory framework may not allow usage of certain variables.

34
00:02:31,400 --> 00:02:34,220
For example, if you decide to build a model.

35
00:02:35,200 --> 00:02:37,720
To identify credit worthiness of a customer.

36
00:02:38,130 --> 00:02:39,760
This is the profile of a customer.

37
00:02:40,600 --> 00:02:43,140
And your model sees that a person's gender.

38
00:02:44,230 --> 00:02:46,540
All religion is a significant variable.

39
00:02:47,740 --> 00:02:50,680
Now, if you base your decisions on this model.

40
00:02:51,850 --> 00:02:56,350
You will be treating people of different genders or different religions differently.

41
00:02:57,340 --> 00:03:00,760
This situation can be considered as one of discrimination.

42
00:03:01,300 --> 00:03:06,610
And if you cannot base your decision on a particular variable, that is, if you will not accept the

43
00:03:06,610 --> 00:03:09,070
result of your model basis, a particular variable.

44
00:03:09,550 --> 00:03:11,260
There is no point keeping that variable.

45
00:03:13,110 --> 00:03:17,160
So keep this in mind when selecting sensitive variables for analysis.

46
00:03:18,980 --> 00:03:23,300
Last point is, again, trading the importance of business knowledge.

47
00:03:24,260 --> 00:03:27,260
Don't just take the variable because you have data available for it.

48
00:03:28,160 --> 00:03:33,590
I mean, yes, we can do exploratory analysis where without business knowledge, we try to identify

49
00:03:33,590 --> 00:03:34,160
a pattern.

50
00:03:34,820 --> 00:03:40,430
But when we are establishing a cause and effect relationship, try to keep only variables that make

51
00:03:40,430 --> 00:03:41,310
logical sense.

52
00:03:42,660 --> 00:03:48,180
To complement your understanding whether the relationship is related or not, you should use the by

53
00:03:48,180 --> 00:03:49,170
various analysis.

54
00:03:50,820 --> 00:03:56,810
If you think house prices will increase with better air quality, that should be reflected in these

55
00:03:56,820 --> 00:03:57,570
scatterplot.

56
00:03:58,400 --> 00:04:00,180
We saw scatterplot of crime did.

57
00:04:00,510 --> 00:04:05,990
It was non-linear, but still it had a recognizable pattern indicating some type of relationship.

58
00:04:07,310 --> 00:04:13,370
But when we plotted rainfall, it was completely, uniformly distributed regardless of race.

59
00:04:14,640 --> 00:04:20,550
This is telling us that rainfall may not be having a significant impact on the outward variable.

60
00:04:21,610 --> 00:04:24,640
So we have an option to delete these rainfall variable.

61
00:04:24,710 --> 00:04:25,150
Also.

62
00:04:26,540 --> 00:04:34,610
As you have seen, identifying, adding and removing variables is an iterative process, and we'll be

63
00:04:34,610 --> 00:04:37,280
doing it even post regression analysis.

64
00:04:38,880 --> 00:04:43,920
We will remove the variables we have identified using these software tools from our dataset.
