WEBVTT 00:01.380 --> 00:09.270 I don't know if we have discussed about the supervised learning models, we have discussed about the 00:09.270 --> 00:18.630 linear regression and they have understood why aggression is used, meaning aggression is used for finding 00:18.630 --> 00:20.190 out continuous values. 00:20.820 --> 00:28.440 Now I want to find out something like height, weight or Cecile's price. 00:28.650 --> 00:32.340 In those situations, I can easily use linear regression. 00:33.210 --> 00:41.250 But now we have another type of problem, which is logistic problem, which is a problem where we want 00:41.250 --> 00:43.770 to find both different classes. 00:44.460 --> 00:49.890 We want to find out if a particular value is yes or no. 00:50.110 --> 00:53.250 We want to find out answer to a problem. 00:53.490 --> 01:02.190 That is, if someone will buy my product or not, whether someone should default on a loan or not. 01:02.550 --> 01:09.260 So these are different type of problems which we will be solving on in case of logistic regression. 01:09.900 --> 01:14.250 But why is it called regression when we are not really doing regression? 01:14.820 --> 01:15.290 Right. 01:15.530 --> 01:22.140 Regression means that we want to predict continuous values, but in logistic regression we are doing 01:22.140 --> 01:23.070 classification. 01:23.070 --> 01:28.170 That is, we are trying to find out the answer to a class that is a categorical variable. 01:28.440 --> 01:35.700 That is, we want to find out if someone is a particular animal, it is a cat or not, or if someone 01:35.700 --> 01:44.280 will default on a loan or not, if someone will buy my product or not solve some problem which has us 01:44.580 --> 01:46.630 a high class problem. 01:47.490 --> 01:53.610 So these type of problems should not be considered like a regression problem. 01:53.610 --> 01:54.000 Right. 01:54.180 --> 01:56.890 But still, we are calling it a regression problem. 01:57.090 --> 02:01.980 Now, this is because not because it is a regression algorithm. 02:01.980 --> 02:04.800 It is classification algorithm only. 02:04.980 --> 02:14.310 But the problem is that the logistic regression is using regression in donelli in the calculations, 02:15.090 --> 02:18.660 and only it is performing some kind of regression. 02:18.960 --> 02:23.500 That is the reason why we are calling it logistic regression back. 02:23.700 --> 02:32.100 Logistic regression is used for classification problems while for continuous value problems for regression 02:32.100 --> 02:32.730 problems. 02:32.970 --> 02:35.170 We use linear regression. 02:36.000 --> 02:39.960 So this is something which you need to keep in mind always. 02:41.570 --> 02:47.860 Now, you have seen how the model continues, medic responds with limited aggression technique, but 02:47.870 --> 02:54.860 in many business scenarios, our target is binary in nature by any means, something which has the answer, 02:54.870 --> 02:56.000 zero or one. 02:56.600 --> 03:02.760 So, for example, whether someone will buy my product, whether someone will default on a loan or not. 03:03.050 --> 03:09.380 So answer to all of these question and many other such questions is yes or no. 03:10.130 --> 03:18.290 We can convert that to zero and one and then try to model them with linear regression technique, but 03:18.290 --> 03:20.790 that does not give good result. 03:21.680 --> 03:23.000 So let's have a look. 03:25.130 --> 03:33.320 So let us have a look at this particular image here now in this particular image, we have the data 03:33.320 --> 03:34.400 point like this. 03:34.640 --> 03:37.730 That is one of the barometer. 03:37.730 --> 03:43.010 One of the target value is zettl and another target value is one. 03:43.760 --> 03:51.620 Now, in this case, if I will apply a linear regression and I will try to put a line here, then the 03:51.620 --> 04:00.410 line will not be able to satisfy the problem because it is going diagonally from the data on my glasses 04:00.410 --> 04:01.370 on like this. 04:01.550 --> 04:05.540 So how will I make my decisions with such kind of a line? 04:07.070 --> 04:14.690 So you can see how far badly this will feel, and that is why we really need something like this, so 04:15.380 --> 04:23.390 what we need is we will create something like this, the image which is present Velu. 04:25.230 --> 04:31.980 Now, to actually reach to this point, to actually reach to this kind of construct, which will give 04:31.980 --> 04:40.470 us this kind of line, this kind of goal, which will be able to map to this data, which is in a class 04:40.470 --> 04:45.090 form, we will have to develop certain kind of understanding. 04:46.010 --> 04:53.640 So then now we have seen that regression approach will feel at different levels. 04:53.840 --> 04:58.670 So we will have to look at these kind of problems from a different perspective. 04:59.060 --> 05:06.110 So let us consider the particular data where we have different values for yes, no, yes, no. 05:06.230 --> 05:07.910 And different age values. 05:10.710 --> 05:19.500 So in this condition for this particular type of data, here we were we ask the children of various 05:19.500 --> 05:23.400 ages whether they are afraid of ghost or not. 05:24.150 --> 05:27.570 Now, here we have different answers from these children. 05:27.780 --> 05:34.210 So the children of age four, there are four children who say yes, one child who says no. 05:34.530 --> 05:39.270 There are three children of age five who say yes to say no. 05:39.510 --> 05:45.000 Six of three to children of age six who say yes and so on. 05:45.510 --> 05:47.250 So now what we will do. 05:48.320 --> 05:56.060 Now, if someone will ask a question that what might be the response for a child who's ages seven, 05:56.210 --> 06:04.010 then you can see that the response will be no, because here the majority of the children of age seven 06:04.190 --> 06:07.400 have responded with no right. 06:07.640 --> 06:11.150 So what did you do there to find out? 06:11.400 --> 06:13.910 You found out the probability of it. 06:14.780 --> 06:21.860 You would just simply check and the probability of the response being low when the age is seven and 06:21.860 --> 06:25.310 you naturally guessed that the answer will be no. 06:25.880 --> 06:33.380 So in this case, we did not model why we did not find out the value. 06:33.380 --> 06:36.800 We simply create a model for the probability. 06:37.670 --> 06:44.360 So in this case, what we will be doing is we will simply create the model with probability of Vike 06:44.360 --> 06:44.720 will do. 06:44.720 --> 06:45.140 Yes. 06:45.320 --> 06:48.800 Or probability of equal to no. 06:50.700 --> 06:59.370 So for this, we will be leaders of the north, this entire thing, by being the probability so we can 06:59.370 --> 07:06.650 see visible to be done on, plus be the one with the X one, plus BWB likes to be WB Deakes. 07:07.050 --> 07:10.680 So these are the equation which we used to have for integration. 07:10.980 --> 07:13.940 But now we are not finding out the Nenad equation. 07:13.950 --> 07:19.440 We are not actually finding out the continuous value, but we are finding out the probability. 07:19.450 --> 07:24.270 So we will say that this should be the probability of this equation, this probability. 07:24.270 --> 07:26.880 We will assign it as the probability. 07:28.650 --> 07:35.610 Now, this will be a little problematic because the probability value has to range from zero to one, 07:35.910 --> 07:42.120 while the regression value of this particular equation value will range from minus infinity to plus 07:42.120 --> 07:42.720 infinity. 07:42.990 --> 07:49.110 Now, for this, we will have to transform this particular equation that must be done on. 07:49.130 --> 07:56.100 Plus, with our next month, as we do X plus, we get to expand so on into a particular range where 07:56.100 --> 08:03.030 this will range from zero to one, then we can use this equation to find out the probability. 08:04.970 --> 08:06.440 Now, let us go back. 08:11.180 --> 08:14.220 Now, let us see what are we trying to predict here? 08:15.050 --> 08:18.260 Now the classes which are being predicted. 08:18.290 --> 08:23.640 So here we are trying to predict different classes and we are not predicting continuous numbers. 08:23.640 --> 08:25.010 So that does not target. 08:25.070 --> 08:28.180 We are not solving the problem. 08:28.430 --> 08:36.710 The name only has logistic regression, but the main purpose here is to solve a classification problem. 08:36.980 --> 08:43.100 So what are different types of classes, which would be the different type of classes for the dog or 08:43.100 --> 08:45.270 dog bus or field? 08:45.530 --> 08:48.600 Yes or no approval or rejection. 08:48.710 --> 08:53.780 So these are these two types of values which could be placed in the logistic regression problem. 08:55.030 --> 09:04.480 Now, these two classes should be of a type which are more possible together, like someone cannot force 09:04.480 --> 09:06.630 and feel the exam the same thing. 09:06.880 --> 09:09.960 So we cannot have something which can occur together. 09:10.260 --> 09:14.460 An animal cannot be OK and go vote together. 09:14.470 --> 09:15.310 I'm the same thing. 09:15.350 --> 09:21.720 It has to be either a gag order, but a person has to either pass the exam or fail exam. 09:22.000 --> 09:22.320 Right. 09:22.600 --> 09:31.180 So now the probability of the outcome being in favor of one class as finding out just how class may 09:31.180 --> 09:34.190 not be enough to clarify the understanding. 09:34.360 --> 09:36.160 So let's see if I. 09:37.340 --> 09:48.200 Simply find out the hard place that is something is a gap or something is a dog that can be a feasible 09:48.200 --> 09:52.370 solution because I want to know more about. 09:54.130 --> 10:02.200 Simply knowing that a person let us see, we have an example here that we are trying to predict whether 10:02.200 --> 10:07.830 a particular person might have a heart disease or not for their particular ages. 10:08.260 --> 10:18.340 Now, if they say yes for both age 45 and 85, that is not informative enough because a person was ages 10:18.400 --> 10:25.870 45 will have less probability of having a heart disease in comparison to person whose age is eighty 10:25.870 --> 10:32.110 five, the person whose ages 85 is obviously more at risk of having a heart disease. 10:33.390 --> 10:40.530 So for solving such problems, we actually need to know the probability of something happening if we 10:40.530 --> 10:47.070 have a patient, a cancer patient, a person who is destined for cancer, and I simply say that you 10:47.070 --> 10:47.860 have cancer. 10:47.880 --> 10:50.590 Yes or no, that is not informative enough. 10:50.880 --> 10:54.650 I want to know, what are the chances of someone's death? 10:54.660 --> 10:56.310 Would it be considered normal? 10:56.640 --> 11:04.140 So I need to know if there is a 90 percent chance of someone having cancer or there is a 10 percent 11:04.140 --> 11:05.860 chance of someone having cancer. 11:06.090 --> 11:10.090 And based on that, I can actually prescribe the next step. 11:10.370 --> 11:16.030 That is what kind of test that person should have or what kind of medicines that person should have. 11:16.260 --> 11:20.870 So those things will be dependent on the probability, which I will be getting Golia. 11:22.160 --> 11:22.460 Right. 11:22.850 --> 11:33.230 So this implies that we want to find out the probability of the value being one given access to that 11:33.230 --> 11:41.990 is we want to find out probability of an animal being gagged, given that it has certain characteristics. 11:42.540 --> 11:49.280 For example, the characteristics would be the animal, says M. or the animal has whiskers on the face. 11:49.310 --> 11:52.670 So these are different kind of characteristics which we have. 11:54.130 --> 12:03.940 So we want to find out we want to predict the probability of vibin one given X, so what are the difference 12:03.940 --> 12:06.860 between linear regression and logistic regression? 12:07.120 --> 12:13.540 The difference is that of the target value for one that is in linear regression. 12:13.550 --> 12:19.810 We want to find out the continuous dependent variable, the value which we are trying to predict is 12:19.810 --> 12:21.820 a continuous dependent variable. 12:23.030 --> 12:32.810 In case of logistic regression, we want to find out about certain classes that is either a value zero 12:32.810 --> 12:38.000 or one based on the probability value, which will be ranging from zero to one. 12:38.210 --> 12:45.050 All that is we are trying to find out a value which is between range zero and one, while in case of 12:45.050 --> 12:51.020 linear regression, we want to find out the value, which is ranging from minus infinity to plus infinity, 12:51.020 --> 12:57.110 because cantinas value will be ranging from minus infinity to plus infinity, while the probabilities, 12:57.440 --> 13:01.160 the probability ranges from zero to one. 13:02.710 --> 13:08.230 Now, what are different, but we will see just for linear regression, the attributes are independent 13:08.230 --> 13:08.880 variables. 13:08.890 --> 13:11.730 These can be continuous or discrete in nature. 13:12.130 --> 13:19.060 And in case of logistic regression, the again, the variables can be continuous or discrete. 13:19.090 --> 13:22.090 So there is a similar kind of you just can be present in both. 13:23.200 --> 13:28.510 Now, what is the equation that we are forming in case of media regression, the equation which is formed 13:28.510 --> 13:37.720 in case of regression, is in the form visible to be done, or plus B, the one in two X one plus some 13:37.720 --> 13:46.630 particular added value by in case of logistic regression, the equation is probability is equal to one 13:46.630 --> 13:53.950 upon one plus E to the power minus the same equation for linear equation be done or plus B that takes 13:53.950 --> 13:56.070 one plus B that we do and so on. 13:56.620 --> 14:02.070 So how we form this equation is what we will learn in this particular session. 14:02.410 --> 14:06.880 But for now, you can just know that this is the equation for logistic regression. 14:08.150 --> 14:13.810 Now, what are the different type of problems which we solve in case of linear regression, these are 14:14.020 --> 14:16.010 temperature marks of student. 14:16.310 --> 14:17.980 What are the future scenes? 14:18.170 --> 14:20.720 What are the prices of commodities? 14:21.260 --> 14:26.960 These are different things which we're finding out in case of linear regression, in case of logistic 14:26.960 --> 14:33.230 regression, what we're finding out in days of logistic regression, we are finding of the binary classes. 14:33.440 --> 14:36.860 That is if if an animal is cat or dog. 14:37.870 --> 14:43.940 If someone passes or fails the examination, if the loan would be approved or rejected. 14:44.140 --> 14:49.450 So these are different kind of things which we predict in case of logistic regression. 14:49.990 --> 14:51.190 Now, let's go for the. 14:52.410 --> 14:55.950 Now, what is the definition of logistic regression? 14:57.580 --> 15:06.250 The logistic regression technique is used for finding out results which can be presented in the binary 15:06.250 --> 15:16.150 zero or one or two or four or more yes or no values means that the outcome could only be in either one 15:16.150 --> 15:18.010 form or of the two. 15:18.820 --> 15:26.960 For example, it can be utilized when we need to find the probability of success or failure of an event. 15:28.480 --> 15:35.800 So here we are finding out if a particular animal is a gag order dog, and here we are finding out the 15:35.800 --> 15:41.200 values from either buy is equal to zero or V. is equal to one. 15:43.280 --> 15:51.500 This we have already seen that this is the linear equation, which gives us the continuous values which 15:51.500 --> 15:58.400 would which would not be of much use if we want to classify Target into two different classes because 15:58.400 --> 16:01.220 the Nenad equation will not be able to map these values. 16:01.820 --> 16:08.660 So to map linear regression equation such that we could get above as shown in figure to. 16:10.230 --> 16:12.630 Which could give us values. 16:13.680 --> 16:17.250 Close to zero and one for a given attribute. 16:17.560 --> 16:19.610 So that is our objective here. 16:21.670 --> 16:29.770 So why is it called aggression, if it is finding out biology classes, so it is underlying Dick is 16:29.770 --> 16:32.120 quite similar to linear regression. 16:32.440 --> 16:39.160 So as we know, the equation which we will be creating would be kind of using the linear regression 16:39.160 --> 16:43.950 equation, but only the Wii will be switched with the probability value. 16:45.760 --> 16:49.870 So that logic is taken from the logic function. 16:49.900 --> 16:56.530 This is the logic function is one upon one plus E to the power minus the. 16:57.710 --> 17:06.090 So that that is used in this classification, so we will be using this logic function for this classification. 17:06.380 --> 17:07.700 We will be converting. 17:08.920 --> 17:17.530 This is equal to be not to be done or the speed of the next one plus two weeks to still be three extreme 17:18.040 --> 17:18.290 two. 17:19.490 --> 17:28.030 This kind of equation, that is the logit equation, that is one point, one plus or minus mid-north 17:28.030 --> 17:30.440 authors with our next one plus 21. 17:32.320 --> 17:39.190 So our task here is to convert this linear model into the logistic model. 17:43.820 --> 17:52.640 So there is this log function which will take care of the range mismatch issue, how this happens, 17:52.640 --> 18:00.410 we will have a look at it in the next session where we will study about the mathematics behind the logistic 18:00.410 --> 18:01.010 regression. 18:01.400 --> 18:03.490 So let's go to the next session.