WEBVTT 00:01.360 --> 00:05.470 In this session, we will discuss about support washing vector machines. 00:07.660 --> 00:09.190 Support vector machines. 00:11.850 --> 00:16.350 Ah, one of the most popular supervised machine learning algorithms. 00:17.370 --> 00:24.960 And it is used for classification as well as aggression problems, but usually it is used for the classification 00:24.960 --> 00:25.530 problems. 00:26.660 --> 00:34.400 The goal of the support vector machine algorithm is to create the best line or decision boundary that 00:34.400 --> 00:37.760 can segregate the dimensional space into classes. 00:39.260 --> 00:46.880 So we will be creating several lines between these two classes so that we are able to segregate them 00:47.360 --> 00:49.040 into different classes. 00:51.140 --> 01:01.100 Now, these lines will be helpful, as we can easily put the new data in the correct category using 01:01.100 --> 01:10.810 these lines of segregation and this decision boundary, this line which we have created is called hyperbole. 01:12.230 --> 01:20.990 Now, because this is just a day data, so this is a lie, but when it becomes a multiple dimensional 01:20.990 --> 01:24.870 data, because we will be having a lot of features which we will be dealing with. 01:25.220 --> 01:30.950 So in that case, it will become a hyperbole because it will be not multi-dimensional space. 01:32.910 --> 01:40.880 Now, for SVM, we choose the extreme point or vectors that help in creating the hyperbole. 01:41.730 --> 01:45.390 Now, these extreme cases are called support vectors. 01:46.080 --> 01:55.170 So if we choose in these two classes, so for these two classes, the two points which are nearest to 01:55.170 --> 02:05.370 this particular class on this and this particular triangle, and the two point from this class which 02:05.370 --> 02:10.140 are nearest to the triangle class are this and this point. 02:11.600 --> 02:19.700 So from these points, which are nearest to the other class, these are called support victims. 02:20.630 --> 02:26.450 So these two points will be known as support weapons for each class. 02:27.820 --> 02:34.690 And hence, that is where this algorithm is called support vector machines, because it is using these 02:34.690 --> 02:39.320 support vectors to actually create this line or type of plane. 02:41.170 --> 02:44.950 So let us see what are different shapes which we have. 02:46.980 --> 02:52.530 So the system can be in linear form also or in non-linear formula. 02:52.640 --> 02:56.000 So now let's have a look at this particular data. 02:56.910 --> 03:04.320 So here we have this data, which is present in a circular form, so I cannot really draw a single line 03:04.320 --> 03:09.030 here to divide these data points into two different classes. 03:10.120 --> 03:17.620 So what I will have to do is I will have to transform this data from two dimensional to three dimensions. 03:19.270 --> 03:26.560 So when I have a look at this data into another dimension, which is I which I have, I did so here 03:26.560 --> 03:28.290 I have added a dimension Z. 03:28.540 --> 03:34.570 So when I look at this data with respect to, say, the next dimension, I can see that there is one 03:34.570 --> 03:37.060 line of separation between both of the classes. 03:38.630 --> 03:48.050 So I will play I will place a hybrid plane between both of these dimensions and that type of plane will 03:48.050 --> 03:52.220 actually help us to create a line of segregation between the. 03:53.540 --> 03:59.720 So here you can see that on transforming back to the spy plane, the leanness apparatus has become a 03:59.990 --> 04:03.850 circle and hence it was able to segregate the classes. 04:08.080 --> 04:18.070 Now, let us have a look at these SBM, so what SVM does is it finds the most similar examples between 04:18.070 --> 04:18.770 the classes. 04:20.110 --> 04:27.610 So these examples will be the support this, so when we're looking at this particular class, which 04:27.610 --> 04:34.310 is the mango class, it tries to find out an apple, which is more like and mango. 04:34.930 --> 04:39.950 So mangoes are usually yellow and longer in size, oval in size. 04:40.150 --> 04:46.720 So we look for an apple, which is a little bit a little yellowish in color and a little longer than 04:46.720 --> 04:49.440 usual, or apples. 04:50.140 --> 04:53.950 So those become the support, therefore, the apples. 04:55.370 --> 04:59.270 Then when they look at the apple does. 05:00.430 --> 05:08.290 They tried to find out the which just a little shorter in height, not really oval, but circular in 05:08.290 --> 05:12.430 shape, and they found them as these support victims. 05:13.760 --> 05:21.820 Now, between these support vectors, we draw the line, these support vectors actually connected to 05:21.860 --> 05:30.500 the line of segregation, and between these two lines, we create one line, which is having the maximum 05:30.500 --> 05:32.990 margin from these two lines. 05:34.700 --> 05:42.290 And this margin is actually called this line, which we have created is actually called the Hyper. 05:46.770 --> 05:53.370 So here's the same thing has been explained, that, for example, in Mango's example, other Al Gore 05:53.640 --> 06:01.840 will try to find the differences between Mango's and that is Mingo's elliptical and yellow by police 06:01.840 --> 06:03.030 around Andric. 06:03.810 --> 06:07.590 But SBM will actually try to find out Manguel. 06:07.590 --> 06:13.440 That looks like an apple that is written down, an apple that actually looks like mineable, which is 06:13.440 --> 06:18.770 yellow and elliptical, and then use these as the support windows. 06:23.090 --> 06:25.170 Now, here you can visualize the same thing. 06:25.490 --> 06:29.600 So we have this red glass and here we have the blue glass. 06:29.960 --> 06:37.330 So we have found the ones which are nearest to the blue one and all of them are supposed to make those. 06:37.340 --> 06:43.670 And here the blue ones, which are nearer to the right are the subject victims from the blue glass and 06:43.670 --> 06:52.800 from them, we have found out the maximum margin and maximizing the margin on both the directions we 06:52.850 --> 06:53.640 create of. 06:54.830 --> 07:03.500 Separating hyperbole now, once we have decided upon these supposed victories, we don't really need 07:03.530 --> 07:05.600 the other points from this data. 07:06.800 --> 07:13.610 Now, because we don't really need these other points from the data, so we don't really need to compare 07:13.610 --> 07:16.820 the other points while classifying world. 07:17.920 --> 07:25.210 All we need is this hyper plane, once this hyper plane has been generated, whatever point we are looking 07:25.210 --> 07:32.170 at, we will just compare the point with the hyperbole and the direction in which the point lies from 07:32.170 --> 07:35.080 the hydroplaned and decide what class it belongs to. 07:37.970 --> 07:47.660 Now, let us look at the steps so we select in case we have a data, then we select all hyper planes 07:47.960 --> 07:51.730 with separate the data with no points between them. 07:51.860 --> 07:53.270 That is the red line. 07:53.690 --> 07:55.700 This these lines. 07:57.410 --> 08:01.200 These red lines, which do not have any data in between. 08:01.490 --> 08:04.910 So we create these two hyperlinked. 08:06.430 --> 08:14.140 Now we maximise their distance, that is the margin between them is maximized once we have maximized 08:14.140 --> 08:21.730 these distances, then the average line that is the line between both of those two red lines will be 08:21.730 --> 08:23.320 the decision boundary. 08:24.220 --> 08:28.270 The line in between the green line is called the decision boundary. 08:28.570 --> 08:37.090 We could have had multiple lines in between like the line or line B, but because it has the maximum 08:37.090 --> 08:42.910 distance from both line and B, hence we use the line E as the hyper.