1 00:00:01,400 --> 00:00:08,700 In this video, we will learn about confusion matrix after we have trained our model and our model is 2 00:00:08,700 --> 00:00:10,330 predicting the losses of life. 3 00:00:11,760 --> 00:00:17,640 We can make a matrix called confusion matrix, which can tell us the performance of our model. 4 00:00:20,150 --> 00:00:23,120 The Matrix is very simple in the columns. 5 00:00:23,570 --> 00:00:25,430 We put the true value of the glasses. 6 00:00:26,970 --> 00:00:30,770 Like yes or no or defaulted or not defaulted. 7 00:00:32,860 --> 00:00:37,220 In the news, we put the predicted value of response to this. 8 00:00:37,410 --> 00:00:39,750 Yes and no is predicted by our model. 9 00:00:40,450 --> 00:00:40,990 This. 10 00:00:41,160 --> 00:00:41,910 Yes and no. 11 00:00:42,090 --> 00:00:45,340 Is the value which we already have in the training sector. 12 00:00:46,830 --> 00:00:51,360 So not this cell represents where the actual value is. 13 00:00:51,360 --> 00:00:51,720 No. 14 00:00:52,110 --> 00:00:53,820 And when we predicted no. 15 00:00:54,630 --> 00:00:57,360 Therefore, this is a correct prediction by our model. 16 00:00:58,620 --> 00:01:03,570 Similarly, this cell is giving us the cone of cases where actual value is. 17 00:01:03,570 --> 00:01:03,930 Yes. 18 00:01:04,140 --> 00:01:04,890 And we predicted. 19 00:01:04,890 --> 00:01:05,250 Yes. 20 00:01:05,760 --> 00:01:07,530 So this is also correct predictions. 21 00:01:08,400 --> 00:01:10,530 These two cells are giving correct predictions. 22 00:01:11,010 --> 00:01:14,880 And these two cells are giving us wrong predictions here. 23 00:01:15,150 --> 00:01:16,820 The true value was yes. 24 00:01:17,580 --> 00:01:19,400 Whereas we predicted no. 25 00:01:20,340 --> 00:01:21,690 It truly was no. 26 00:01:21,930 --> 00:01:23,010 We predicted yes. 27 00:01:25,260 --> 00:01:27,460 These two errors are guard type one. 28 00:01:27,600 --> 00:01:28,680 And I do it at. 29 00:01:31,610 --> 00:01:33,770 Type one error is when you predict. 30 00:01:33,800 --> 00:01:36,690 Yes, but the actual value is no. 31 00:01:38,630 --> 00:01:41,570 And type two error is when we predict. 32 00:01:41,600 --> 00:01:41,990 No. 33 00:01:42,230 --> 00:01:43,250 But the actual values. 34 00:01:43,250 --> 00:01:43,690 Yes. 35 00:01:45,380 --> 00:01:48,210 It is easy to get confused between type one and type two. 36 00:01:48,240 --> 00:01:48,380 It. 37 00:01:49,820 --> 00:01:51,200 I saw this image somewhere. 38 00:01:51,500 --> 00:01:52,680 Just help me remember. 39 00:01:52,880 --> 00:01:54,320 Which is type one error. 40 00:01:54,410 --> 00:01:55,650 And what is they do it. 41 00:01:57,510 --> 00:02:05,210 They when it it is when are immortalizing that this person is pregnant when that person is not or cannot 42 00:02:05,210 --> 00:02:05,540 be. 43 00:02:07,090 --> 00:02:07,850 I do it. 44 00:02:07,850 --> 00:02:13,250 It is when I immortalizing that this lady is not pregnant when she actually is. 45 00:02:14,210 --> 00:02:21,290 So using this image you can easily remember that type one error is a false positive and type two error 46 00:02:21,350 --> 00:02:22,640 is a false negative. 47 00:02:24,680 --> 00:02:31,670 The point of segregating these two type of error is that often the cost of making each type of error 48 00:02:31,760 --> 00:02:32,380 is different. 49 00:02:33,690 --> 00:02:42,690 For example, if you have a safe full of money, which opens when it detects your face, type one error. 50 00:02:42,870 --> 00:02:52,440 In that case will be if the safe opens up when it is someone else instead of you type two, it is that 51 00:02:52,540 --> 00:02:55,290 you are standing in front of it and it is not opening. 52 00:02:57,360 --> 00:03:02,360 Now, you may accept I do it at that maybe two out of 10 times. 53 00:03:02,490 --> 00:03:03,720 It does not open for you. 54 00:03:04,470 --> 00:03:08,340 You can always replay in such a scenario, but do not accept. 55 00:03:08,640 --> 00:03:16,850 If it opens up for any two out of ten random people, that's not really safe in such a scenario. 56 00:03:17,380 --> 00:03:19,200 The cost of edit is different. 57 00:03:19,650 --> 00:03:23,460 We often shift bond devalue to a more conservative value. 58 00:03:24,570 --> 00:03:30,780 So, for example, your staff will not open when it is more than 50 percent sure, but it will only 59 00:03:30,780 --> 00:03:33,840 open when it has more than 80 percent sure that it is you. 60 00:03:36,450 --> 00:03:42,500 So confusion matrix is a common way to show how well your predictions fare against the two values. 61 00:03:43,250 --> 00:03:47,330 And we are going to learn how to create confusion, metrics and this software package.