1
00:00:00,600 --> 00:00:04,220
In this section, we are going to cover moving average smoothly.

2
00:00:04,980 --> 00:00:10,680
You got a glimpse of this when we saw the window type features in the feature engineering lecture.

3
00:00:12,270 --> 00:00:15,360
Now, we will cover this in a more structured format.

4
00:00:17,340 --> 00:00:20,100
So we will cover these three things first.

5
00:00:20,610 --> 00:00:24,450
What is moving average and why we do it?

6
00:00:25,070 --> 00:00:31,020
Then we will see these two types of moving averages centered, window moving average and grilling window

7
00:00:31,020 --> 00:00:31,770
moving average.

8
00:00:33,770 --> 00:00:40,760
Lastly, we will see how moving average is used for feature engineering and how it is used, what prediction

9
00:00:40,760 --> 00:00:41,330
purposes.

10
00:00:42,050 --> 00:00:42,770
So let's start.

11
00:00:45,230 --> 00:00:53,330
Moving out, it's moving is simply creating a new series where the values are averages of the draw observations

12
00:00:54,080 --> 00:00:55,520
in the original time cities.

13
00:00:57,050 --> 00:01:01,230
Let us see an example to understand this here.

14
00:01:01,730 --> 00:01:06,380
I have Monterey's average temperature data in the month of January.

15
00:01:06,860 --> 00:01:10,030
We had average temperature of 39 degrees Fahrenheit.

16
00:01:11,450 --> 00:01:14,660
In fact, it was 42 in the month of March.

17
00:01:14,930 --> 00:01:16,670
It was 50 degrees Fahrenheit.

18
00:01:18,200 --> 00:01:23,390
Now, in March, we can find out the average of last three months.

19
00:01:24,620 --> 00:01:28,880
The average temperature of last three months would be 44 degrees Fahrenheit.

20
00:01:30,890 --> 00:01:33,110
We can do this for other months.

21
00:01:33,350 --> 00:01:35,130
And this is also true.

22
00:01:35,420 --> 00:01:41,400
If we have April, we can find the average of February, March and April.

23
00:01:43,250 --> 00:01:49,520
In the month of May, we can average the of March, April and May.

24
00:01:49,850 --> 00:01:54,050
This value comes out to be sixty in this way.

25
00:01:54,650 --> 00:02:00,680
We are averaging the last three values of temperature and creating a new series.

26
00:02:02,140 --> 00:02:09,410
The three values which we are averaging are moving along the cities as we are moving forward in the

27
00:02:09,410 --> 00:02:19,280
cities and the new cities that we get by finding the average of last few months is the new moving average

28
00:02:19,280 --> 00:02:19,790
cities.

29
00:02:21,290 --> 00:02:22,670
But why did we do this?

30
00:02:25,040 --> 00:02:29,300
Let's compare a graph of raw observations and moving averages.

31
00:02:31,700 --> 00:02:37,500
Here you can see that the moving average graph of stock prices is more smooth.

32
00:02:37,820 --> 00:02:38,770
Then the raw data.

33
00:02:41,130 --> 00:02:44,790
It does remove defined green variations between timestamps.

34
00:02:47,010 --> 00:02:47,790
Our agenda.

35
00:02:47,880 --> 00:02:57,030
In doing so is that in this process, we are trying to remove the noise and trying to find the underlying

36
00:02:57,030 --> 00:02:57,660
process.

37
00:02:59,560 --> 00:03:05,050
So you can see in this graph, by averaging out devalues our stock price.

38
00:03:06,610 --> 00:03:12,220
We have identified this underlying process in these stock prices.

39
00:03:15,450 --> 00:03:24,640
Now, in not monthly temperature data, for example, we were averaging last three values, these three

40
00:03:24,640 --> 00:03:29,110
values constitute that window in which we do the averaging.

41
00:03:30,280 --> 00:03:31,960
And this window is moon.

42
00:03:32,230 --> 00:03:40,240
As we move along this series, this window has two parameters that we need to fix before we start doing

43
00:03:40,780 --> 00:03:41,980
moving average modeling.

44
00:03:43,780 --> 00:03:46,110
The first parameter is the window bit.

45
00:03:46,660 --> 00:03:51,980
That is the number of values that we want to average over here.

46
00:03:52,150 --> 00:03:52,690
The window.

47
00:03:52,810 --> 00:03:56,170
It is three because we are averaging three values.

48
00:03:57,400 --> 00:04:01,510
If you want to average last five values window, it becomes five.

49
00:04:04,110 --> 00:04:07,610
The second part of me that is very you want to position this window.

50
00:04:09,990 --> 00:04:17,160
So right now, at the third observation, we position this window so that we average the last three

51
00:04:17,160 --> 00:04:17,730
values.

52
00:04:18,360 --> 00:04:25,020
That is we are averaging the first, second and third value to get the moving average value for third

53
00:04:25,020 --> 00:04:25,620
observation.

54
00:04:27,060 --> 00:04:31,470
So for them, month of March, we got the moving average value of 44.

55
00:04:31,650 --> 00:04:36,210
By averaging temperature values for January, February and March.

56
00:04:38,330 --> 00:04:40,820
This is called really moving average.

57
00:04:42,710 --> 00:04:49,730
On the other hand, if you please, the window in such a way that the current timestep is at its center,

58
00:04:50,750 --> 00:04:53,480
that will be called centered moving average.

59
00:04:55,360 --> 00:05:02,560
So if we find the average of Jan FIB and March and a sign that moving average value to decentered value,

60
00:05:02,830 --> 00:05:07,030
which is February, then this is called centered moving average.

61
00:05:11,520 --> 00:05:16,140
But as you can note, that for centered moving average.

62
00:05:16,380 --> 00:05:20,850
You need to know the next value of the future value.

63
00:05:21,120 --> 00:05:21,630
Already.

64
00:05:23,270 --> 00:05:27,090
And since and discourse, we are trying to predict those future values.

65
00:05:27,390 --> 00:05:30,000
We will not be using these standard moving average.

66
00:05:33,190 --> 00:05:36,400
Now, trailing, moving average can be used in two ways.

67
00:05:38,050 --> 00:05:42,580
One is feature engineering and the other is forecasting.

68
00:05:44,770 --> 00:05:46,840
The context behind both is simple.

69
00:05:47,530 --> 00:05:56,500
If you want to forecast the city's value at time C, D, D, you can use the average value of the previous

70
00:05:56,500 --> 00:05:57,430
few time cities.

71
00:05:58,660 --> 00:06:05,020
If you use these average values as a new feature, then it is feature engineering.

72
00:06:06,790 --> 00:06:13,870
On the other hand, you can simply assign this average value also as the forecasted value for time P

73
00:06:13,870 --> 00:06:14,350
plus one.

74
00:06:16,330 --> 00:06:23,830
So, for example, if you want to forecast for the month of June, you can simply use this moving average

75
00:06:23,830 --> 00:06:24,310
value.

76
00:06:24,760 --> 00:06:28,690
That is sixty has the forecasted value for the month of June.

77
00:06:31,450 --> 00:06:38,860
Of course, this is a very simple and a naive method of forecasting, and it assumes that there is no

78
00:06:38,860 --> 00:06:40,930
trend and seasonality in the data.

79
00:06:42,340 --> 00:06:46,720
So in terms of accuracy, this method does not perform very well.

80
00:06:47,620 --> 00:06:54,580
But in terms of its simplicity and the ease of implementation, it is often preferred to get a rough

81
00:06:54,580 --> 00:06:58,210
estimate of future values using the moving average method.

82
00:07:00,640 --> 00:07:04,290
So that's all the duty we need to know in this particular session.

83
00:07:04,390 --> 00:07:10,810
We will find out how to get billing values and how we use those trailing values as a new feature.