1 00:00:02,060 --> 00:00:08,640 In the last video, we learned about animal models which were able to handle train in the data. 2 00:00:09,750 --> 00:00:17,160 But an obvious shortcoming of remote matter is that it is unable to handle seasonality in the data to 3 00:00:17,160 --> 00:00:18,630 handle seasonality also. 4 00:00:18,900 --> 00:00:22,890 We can use the method of different thing, as we have seen earlier. 5 00:00:23,460 --> 00:00:29,160 We can remove seasonality by subtracting the corresponding value from people's time steps back. 6 00:00:29,790 --> 00:00:35,340 That is, if we have monthly seasonality, we can subtract the previous year. 7 00:00:35,640 --> 00:00:38,790 Same month sales value from this year. 8 00:00:38,970 --> 00:00:40,260 Same month sales value. 9 00:00:42,000 --> 00:00:50,820 This additional different in step is incorporated in a model called Satima, which stands for seasonal 10 00:00:51,030 --> 00:00:54,240 auto regressive integrated moving average method. 11 00:00:54,810 --> 00:00:56,400 Basically seasonal rhema. 12 00:00:58,990 --> 00:01:06,310 Let's look at the parameters of Saudi model, you might be expecting that with seasonality. 13 00:01:06,700 --> 00:01:11,320 Only one more parameter would be added, which will be the period of seasonality. 14 00:01:12,070 --> 00:01:13,270 But that is not the case. 15 00:01:14,440 --> 00:01:16,690 Seasonality, comfort for new parameters. 16 00:01:17,320 --> 00:01:21,970 These are denoted as capital B, D, Q and small M. 17 00:01:24,210 --> 00:01:25,920 This M is the obvious one. 18 00:01:26,190 --> 00:01:27,570 The period of seasonality. 19 00:01:28,710 --> 00:01:35,430 So for monthly seasonality and will be Cuil for weekly seasonality in daily data and will be saving 20 00:01:37,440 --> 00:01:45,150 Beebee, you will be feeling the impact of seasonality on auto regression, integration and moving average. 21 00:01:46,500 --> 00:01:48,390 Let's see how by using an example. 22 00:01:51,240 --> 00:01:55,680 Suppose this is your data which has bought trend and seasonality. 23 00:01:58,230 --> 00:02:07,290 So to handle Trent, you mentioned these three parameters, BDK, you bought this in places that we 24 00:02:07,290 --> 00:02:16,270 do defending ones because these equal to one in this example to remove the drain on this remaining data. 25 00:02:17,220 --> 00:02:18,420 We pick up to lag. 26 00:02:18,420 --> 00:02:22,830 We loose because being equal to two to build other auto regressive model. 27 00:02:24,570 --> 00:02:30,330 And then we build a moving average model on the last three as it was, because it was equal to three. 28 00:02:32,680 --> 00:02:32,940 No. 29 00:02:33,490 --> 00:02:33,690 In. 30 00:02:34,450 --> 00:02:38,020 Along with these three, we will specify four more parameters. 31 00:02:39,880 --> 00:02:42,140 Capital B, B, Q and smaller. 32 00:02:43,990 --> 00:02:52,150 First, after removing the plane, we will do different thing again on this data for different thing. 33 00:02:52,300 --> 00:02:59,580 We will use lag and value and we will do it once because we have said capitally to one. 34 00:02:59,800 --> 00:03:00,580 In this example. 35 00:03:02,920 --> 00:03:08,240 In the graph that I'm showing you, this is an example in which there is eyerly seasonality. 36 00:03:08,900 --> 00:03:10,870 So here M is equal to Bill. 37 00:03:11,650 --> 00:03:15,350 So we have done different thing with lag. 38 00:03:15,480 --> 00:03:18,500 The L value to BEATTIES analyzed this data. 39 00:03:20,630 --> 00:03:26,690 If you need to do double different thing for the moving seasonality, the capitally will be too. 40 00:03:27,740 --> 00:03:29,150 But that is rarely the case. 41 00:03:30,110 --> 00:03:33,230 Usually capitally is equal to one 42 00:03:35,900 --> 00:03:38,660 after being different thing to remove the seasonality. 43 00:03:38,660 --> 00:03:43,910 But also we will be making the model for this. 44 00:03:44,510 --> 00:03:51,860 We will be using these two values and we'll be using the lag of seasonal values also. 45 00:03:52,550 --> 00:03:56,180 So apart from lag one and lag two values. 46 00:03:57,420 --> 00:03:59,520 We'll be using the lag and value. 47 00:03:59,550 --> 00:04:08,940 Also, the seasonal component is giving only one input that is lagging because capital P is equal to 48 00:04:08,940 --> 00:04:09,210 one. 49 00:04:10,350 --> 00:04:15,990 If Capital B is equal to two, we'll be having two inputs from the seasonality. 50 00:04:16,080 --> 00:04:22,440 That is lag M and lag to him since M is equal to twelve for that example. 51 00:04:22,860 --> 00:04:26,580 It will be led to L value and lag guarantee for the value. 52 00:04:29,280 --> 00:04:37,290 The point of doing this is that when we are predicting sales for the coming month of January, along 53 00:04:37,290 --> 00:04:45,030 with these sales data of November and December, it makes sense to use these sales data of last year, 54 00:04:45,090 --> 00:04:46,460 January sales also. 55 00:04:49,060 --> 00:04:57,840 So this capital p value will tell how many sales values of previous years are to be taken for the auto 56 00:04:57,840 --> 00:04:58,500 regression model. 57 00:05:00,900 --> 00:05:07,080 Similarly, for moving average model, we will take Führer's duals from the previous seasons as well. 58 00:05:08,520 --> 00:05:16,560 So this capital Q will tell how many as a dual values of same season are to be taken from the previous 59 00:05:16,560 --> 00:05:16,950 years. 60 00:05:18,930 --> 00:05:20,170 So that the small P. 61 00:05:20,500 --> 00:05:24,200 Q and capital BD, Q and M. 62 00:05:24,780 --> 00:05:26,080 We will build our study. 63 00:05:26,080 --> 00:05:26,600 My model. 64 00:05:31,400 --> 00:05:38,920 And another, an extension of the emo model is Sa'adi Max model, in which, apart from the normal Satima 65 00:05:38,920 --> 00:05:45,700 model, which is based on the times it is, you also import some exogenous variables. 66 00:05:47,590 --> 00:05:53,230 Exogenous variables will be other variables other than the variable to be predicted. 67 00:05:54,910 --> 00:06:02,920 For example, if you're trying to predict the closing price of a security on the stock market, apart 68 00:06:02,920 --> 00:06:12,250 from using the historical values of closing prices, you might also use variables such as open price, 69 00:06:13,060 --> 00:06:18,070 volume, greater, etc. Such variables are exogenous. 70 00:06:20,530 --> 00:06:28,030 Similarly, for predicting the sales of the coming days apart from using the previous day sales values, 71 00:06:28,810 --> 00:06:36,430 we can use other variables, such as whether the coming day is a holiday or not, whether the coming 72 00:06:36,430 --> 00:06:38,200 days are salary days or not. 73 00:06:39,040 --> 00:06:42,550 Whether we are doing any promotions on those days or not and so on. 74 00:06:43,690 --> 00:06:45,490 Such variables will be exogenous. 75 00:06:45,730 --> 00:06:51,850 In this case, such exogenous variables can also be easily added to our model. 76 00:06:52,960 --> 00:06:58,800 We just need to specify additional parameter which will contain this exogenous dataset. 77 00:06:59,380 --> 00:07:07,150 Only thing that we must note is that in this additional dataset date thing should be kept as an index 78 00:07:07,330 --> 00:07:08,650 and not as a separate column. 79 00:07:10,260 --> 00:07:12,590 However, this part is software specific. 80 00:07:13,940 --> 00:07:20,540 So for Saudi, Max, the only thing that you need to know is that Sa'adi Max model will allow you to 81 00:07:20,540 --> 00:07:25,430 use other variables also and not just rely on the city's data.